tooluniverse 1.0.7__py3-none-any.whl → 1.0.9__py3-none-any.whl

tooluniverse/data/packages/cheminformatics_tools.json

@@ -276,7 +276,7 @@
         "pip": "pip install openchem",
         "conda": "conda install -c conda-forge openchem"
       },
-      "usage_example": "# OpenChem deep learning for drug discovery\n# This demonstrates molecular property prediction concepts\n\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.ensemble import RandomForestRegressor\nfrom sklearn.metrics import mean_squared_error, r2_score\nimport tempfile\nimport os\nfrom collections import defaultdict\n\nprint('OpenChem - Deep Learning for Drug Discovery')\nprint('=' * 45)\n\n# Overview of OpenChem capabilities\nprint('OpenChem Features:')\nprint('• Molecular property prediction (ADMET, bioactivity)')\nprint('• Graph neural networks for molecules')\nprint('• Molecular generation and optimization')\nprint('• Multi-task learning for drug discovery')\nprint('• Molecular descriptors and fingerprints')\nprint('• Integration with PyTorch and other ML frameworks')\n\nprint('\\nSupported Molecular Representations:')\nprint('• SMILES strings')\nprint('• Molecular graphs')\nprint('• 3D conformations')\nprint('• Molecular fingerprints (ECFP, MACCS, etc.)')\nprint('• Physicochemical descriptors')\n\n# Simulate molecular dataset\nprint('\\n=== Molecular Dataset Simulation ===')\n\nnp.random.seed(42)\n\n# Generate synthetic molecular data\nn_molecules = 1000\nprint(f'Generating {n_molecules} synthetic molecules...')\n\n# Molecular descriptors (simplified)\ndescriptor_names = [\n    'molecular_weight', 'logP', 'num_donors', 'num_acceptors',\n    'tpsa', 'num_rotatable_bonds', 'num_aromatic_rings',\n    'formal_charge', 'num_heteroatoms', 'fraction_csp3'\n]\n\n# Generate realistic molecular descriptors\nmolecular_data = []\nfor i in range(n_molecules):\n    # Molecular weight: 150-600 Da (drug-like range)\n    mw = np.random.normal(350, 100)\n    mw = np.clip(mw, 150, 600)\n    \n    # LogP: -2 to 6 (lipophilicity)\n    logP = np.random.normal(2.5, 1.5)\n    logP = np.clip(logP, -2, 6)\n    \n    # Hydrogen bond donors: 0-5\n    donors = np.random.poisson(2)\n    donors = np.clip(donors, 0, 5)\n    \n    # Hydrogen bond acceptors: 0-10\n    acceptors = np.random.poisson(4)\n    acceptors = np.clip(acceptors, 0, 10)\n    \n    # Topological polar surface area: 0-200 Ų\n    tpsa = np.random.gamma(2, 30)\n    tpsa = np.clip(tpsa, 0, 200)\n    \n    # Rotatable bonds: 0-15\n    rot_bonds = np.random.poisson(5)\n    rot_bonds = np.clip(rot_bonds, 0, 15)\n    \n    # Aromatic rings: 0-4\n    aromatic_rings = np.random.poisson(2)\n    aromatic_rings = np.clip(aromatic_rings, 0, 4)\n    \n    # Formal charge: typically 0, sometimes ±1\n    formal_charge = np.random.choice([0, 0, 0, 0, 1, -1])\n    \n    # Heteroatoms: 1-8\n    heteroatoms = np.random.poisson(3) + 1\n    heteroatoms = np.clip(heteroatoms, 1, 8)\n    \n    # Fraction sp3 carbons: 0-1\n    frac_csp3 = np.random.beta(2, 2)\n    \n    molecule = {\n        'id': f'MOL_{i:04d}',\n        'molecular_weight': mw,\n        'logP': logP,\n        'num_donors': donors,\n        'num_acceptors': acceptors,\n        'tpsa': tpsa,\n        'num_rotatable_bonds': rot_bonds,\n        'num_aromatic_rings': aromatic_rings,\n        'formal_charge': formal_charge,\n        'num_heteroatoms': heteroatoms,\n        'fraction_csp3': frac_csp3\n    }\n    \n    molecular_data.append(molecule)\n\n# Convert to DataFrame\ndf = pd.DataFrame(molecular_data)\nprint(f'Generated molecular dataset: {df.shape}')\nprint(f'Descriptors: {len(descriptor_names)}')\n\n# Show basic statistics\nprint('\\nDataset statistics:')\nprint(df[descriptor_names].describe().round(2))\n\n# Generate target properties\nprint('\\n=== Target Property Generation ===')\n\n# Simulate bioactivity (IC50 values)\nprint('Generating bioactivity data (IC50 values)...')\n\ndef calculate_bioactivity(row):\n    \"\"\"Simulate bioactivity based on molecular descriptors\"\"\"\n    # Lipinski's rule of five compliance\n    lipinski_score = 0\n    if row['molecular_weight'] <= 500: lipinski_score += 1\n    if row['logP'] <= 5: lipinski_score += 1\n    if row['num_donors'] <= 5: lipinski_score += 1\n    if row['num_acceptors'] <= 10: lipinski_score += 1\n    \n    # Base activity influenced by Lipinski compliance\n    base_activity = 5.0 + (lipinski_score - 2.0) * 1.5\n    \n    # Additional molecular factors\n    tpsa_factor = -0.01 * max(0, row['tpsa'] - 90)  # Penalty for high TPSA\n    flexibility_factor = -0.1 * max(0, row['num_rotatable_bonds'] - 7)  # Penalty for high flexibility\n    aromatic_factor = 0.3 * min(row['num_aromatic_rings'], 3)  # Bonus for aromatics (up to 3)\n    \n    # Combined activity (pIC50)\n    activity = base_activity + tpsa_factor + flexibility_factor + aromatic_factor\n    \n    # Add some noise\n    activity += np.random.normal(0, 0.5)\n    \n    # Ensure reasonable range (4-9 pIC50)\n    activity = np.clip(activity, 4.0, 9.0)\n    \n    return activity\n\n# Calculate bioactivity\ndf['pIC50'] = df.apply(calculate_bioactivity, axis=1)\n\n# Generate additional properties\nprint('Generating ADMET properties...')\n\n# Solubility (LogS)\ndef calculate_solubility(row):\n    \"\"\"Simulate aqueous solubility\"\"\"\n    # Lipophilicity penalty\n    logP_penalty = -0.5 * max(0, row['logP'] - 2)\n    \n    # Molecular weight penalty\n    mw_penalty = -0.005 * max(0, row['molecular_weight'] - 300)\n    \n    # TPSA bonus (polar surface area helps solubility)\n    tpsa_bonus = 0.01 * min(row['tpsa'], 100)\n    \n    # Base solubility\n    base_solubility = -2.0\n    \n    solubility = base_solubility + logP_penalty + mw_penalty + tpsa_bonus\n    solubility += np.random.normal(0, 0.3)\n    \n    return np.clip(solubility, -6.0, 1.0)\n\ndf['logS'] = df.apply(calculate_solubility, axis=1)\n\n# Permeability (Caco-2)\ndef calculate_permeability(row):\n    \"\"\"Simulate cell permeability\"\"\"\n    # LogP correlation\n    logP_factor = 0.3 * row['logP']\n    \n    # TPSA penalty\n    tpsa_penalty = -0.02 * row['tpsa']\n    \n    # Molecular weight penalty\n    mw_penalty = -0.003 * row['molecular_weight']\n    \n    # Base permeability (log Papp)\n    base_perm = -4.5\n    \n    permeability = base_perm + logP_factor + tpsa_penalty + mw_penalty\n    permeability += np.random.normal(0, 0.4)\n    \n    return np.clip(permeability, -7.0, -3.0)\n\ndf['log_Papp'] = df.apply(calculate_permeability, axis=1)\n\n# Half-life (t1/2)\ndef calculate_half_life(row):\n    \"\"\"Simulate metabolic stability\"\"\"\n    # Molecular complexity factor\n    complexity = row['num_rotatable_bonds'] + row['num_heteroatoms']\n    complexity_factor = -0.05 * complexity\n    \n    # Aromatic stabilization\n    aromatic_factor = 0.1 * row['num_aromatic_rings']\n    \n    # Base half-life (log hours)\n    base_t_half = 1.0\n    \n    t_half = base_t_half + complexity_factor + aromatic_factor\n    t_half += np.random.normal(0, 0.3)\n    \n    return np.clip(t_half, -1.0, 3.0)\n\ndf['log_t_half'] = df.apply(calculate_half_life, axis=1)\n\nprint(f'Generated properties: pIC50, logS, log_Papp, log_t_half')\nprint(f'Property value ranges:')\nfor prop in ['pIC50', 'logS', 'log_Papp', 'log_t_half']:\n    print(f'  {prop}: {df[prop].min():.2f} to {df[prop].max():.2f}')\n\n# Apply drug-likeness filters\nprint('\\n=== Drug-likeness Analysis ===')\n\ndef lipinski_filter(row):\n    \"\"\"Apply Lipinski's Rule of Five\"\"\"\n    violations = 0\n    if row['molecular_weight'] > 500: violations += 1\n    if row['logP'] > 5: violations += 1\n    if row['num_donors'] > 5: violations += 1\n    if row['num_acceptors'] > 10: violations += 1\n    return violations\n\ndef veber_filter(row):\n    \"\"\"Apply Veber's rules for oral bioavailability\"\"\"\n    violations = 0\n    if row['tpsa'] > 140: violations += 1\n    if row['num_rotatable_bonds'] > 10: violations += 1\n    return violations\n\ndf['lipinski_violations'] = df.apply(lipinski_filter, axis=1)\ndf['veber_violations'] = df.apply(veber_filter, axis=1)\ndf['drug_like'] = (df['lipinski_violations'] == 0) & (df['veber_violations'] == 0)\n\nprint(f'Drug-likeness analysis:')\nprint(f'  Lipinski compliant (0 violations): {(df[\"lipinski_violations\"] == 0).sum()}')\nprint(f'  Veber compliant (0 violations): {(df[\"veber_violations\"] == 0).sum()}')\nprint(f'  Overall drug-like: {df[\"drug_like\"].sum()}')\nprint(f'  Drug-like percentage: {df[\"drug_like\"].mean():.1%}')\n\n# Machine learning model training\nprint('\\n=== Machine Learning Model Training ===')\n\n# Prepare features and targets\nfeature_cols = descriptor_names\ntarget_cols = ['pIC50', 'logS', 'log_Papp', 'log_t_half']\n\nX = df[feature_cols].values\ny = df[target_cols].values\n\nprint(f'Feature matrix shape: {X.shape}')\nprint(f'Target matrix shape: {y.shape}')\n\n# Split data\nX_train, X_test, y_train, y_test = train_test_split(\n    X, y, test_size=0.2, random_state=42\n)\n\nprint(f'Training set: {X_train.shape[0]} molecules')\nprint(f'Test set: {X_test.shape[0]} molecules')\n\n# Train models for each property\nmodels = {}\nperformance = {}\n\nfor i, target_name in enumerate(target_cols):\n    print(f'\\nTraining model for {target_name}...')\n    \n    # Random Forest model\n    model = RandomForestRegressor(n_estimators=100, random_state=42)\n    model.fit(X_train, y_train[:, i])\n    \n    # Predictions\n    y_pred_train = model.predict(X_train)\n    y_pred_test = model.predict(X_test)\n    \n    # Performance metrics\n    train_r2 = r2_score(y_train[:, i], y_pred_train)\n    test_r2 = r2_score(y_test[:, i], y_pred_test)\n    train_rmse = np.sqrt(mean_squared_error(y_train[:, i], y_pred_train))\n    test_rmse = np.sqrt(mean_squared_error(y_test[:, i], y_pred_test))\n    \n    models[target_name] = model\n    performance[target_name] = {\n        'train_r2': train_r2,\n        'test_r2': test_r2,\n        'train_rmse': train_rmse,\n        'test_rmse': test_rmse,\n        'predictions': y_pred_test,\n        'true_values': y_test[:, i]\n    }\n    \n    print(f'  Training R²: {train_r2:.3f}')\n    print(f'  Test R²: {test_r2:.3f}')\n    print(f'  Test RMSE: {test_rmse:.3f}')\n\n# Feature importance analysis\nprint('\\n=== Feature Importance Analysis ===')\n\nfeature_importance = pd.DataFrame({\n    'feature': feature_cols,\n    **{target: models[target].feature_importances_ for target in target_cols}\n})\n\nprint('Top features for each property:')\nfor target in target_cols:\n    top_features = feature_importance.nlargest(3, target)\n    print(f'\\n{target}:')\n    for _, row in top_features.iterrows():\n        print(f'  {row[\"feature\"]}: {row[target]:.3f}')\n\n# Virtual screening simulation\nprint('\\n=== Virtual Screening Simulation ===')\n\n# Generate new molecules for screening\nn_screening = 10000\nprint(f'Generating {n_screening} molecules for virtual screening...')\n\nscreening_data = []\nfor i in range(n_screening):\n    # Generate random molecular descriptors\n    mw = np.random.normal(350, 120)\n    mw = np.clip(mw, 100, 700)\n    \n    logP = np.random.normal(2.5, 2.0)\n    logP = np.clip(logP, -3, 7)\n    \n    donors = np.random.poisson(2)\n    donors = np.clip(donors, 0, 8)\n    \n    acceptors = np.random.poisson(4)\n    acceptors = np.clip(acceptors, 0, 15)\n    \n    tpsa = np.random.gamma(2, 35)\n    tpsa = np.clip(tpsa, 0, 250)\n    \n    rot_bonds = np.random.poisson(6)\n    rot_bonds = np.clip(rot_bonds, 0, 20)\n    \n    aromatic_rings = np.random.poisson(2)\n    aromatic_rings = np.clip(aromatic_rings, 0, 5)\n    \n    formal_charge = np.random.choice([0, 0, 0, 0, 1, -1])\n    \n    heteroatoms = np.random.poisson(4) + 1\n    heteroatoms = np.clip(heteroatoms, 1, 12)\n    \n    frac_csp3 = np.random.beta(2, 2)\n    \n    screening_data.append([\n        mw, logP, donors, acceptors, tpsa, rot_bonds,\n        aromatic_rings, formal_charge, heteroatoms, frac_csp3\n    ])\n\nX_screening = np.array(screening_data)\n\n# Predict properties for screening library\nprint('Predicting properties for screening library...')\nscreening_predictions = {}\n\nfor target in target_cols:\n    predictions = models[target].predict(X_screening)\n    screening_predictions[target] = predictions\n\n# Apply filters for hit identification\nprint('\\nApplying screening filters...')\n\n# Create screening DataFrame\nscreening_df = pd.DataFrame(X_screening, columns=feature_cols)\nfor target in target_cols:\n    screening_df[f'pred_{target}'] = screening_predictions[target]\n\n# Apply drug-likeness filters\nscreening_df['lipinski_violations'] = screening_df.apply(lipinski_filter, axis=1)\nscreening_df['veber_violations'] = screening_df.apply(veber_filter, axis=1)\nscreening_df['drug_like'] = (\n    (screening_df['lipinski_violations'] == 0) & \n    (screening_df['veber_violations'] == 0)\n)\n\n# Activity filters\nactivity_threshold = 6.5  # pIC50 > 6.5 (IC50 < 316 nM)\nsolubility_threshold = -4.0  # logS > -4 (> 0.1 mM)\npermeability_threshold = -5.5  # log Papp > -5.5\n\nhits = screening_df[\n    (screening_df['pred_pIC50'] > activity_threshold) &\n    (screening_df['pred_logS'] > solubility_threshold) &\n    (screening_df['pred_log_Papp'] > permeability_threshold) &\n    (screening_df['drug_like'] == True)\n]\n\nprint(f'Screening results:')\nprint(f'  Total molecules: {len(screening_df):,}')\nprint(f'  Drug-like molecules: {screening_df[\"drug_like\"].sum():,}')\nprint(f'  Active hits (pIC50 > {activity_threshold}): {(screening_df[\"pred_pIC50\"] > activity_threshold).sum():,}')\nprint(f'  Final hits (all criteria): {len(hits):,}')\nprint(f'  Hit rate: {len(hits) / len(screening_df):.1%}')\n\nif len(hits) > 0:\n    print(f'\\nTop 5 hits:')\n    top_hits = hits.nlargest(5, 'pred_pIC50')\n    for idx, hit in top_hits.iterrows():\n        print(f'  Hit {idx}: pIC50={hit[\"pred_pIC50\"]:.2f}, '\n              f'logS={hit[\"pred_logS\"]:.2f}, MW={hit[\"molecular_weight\"]:.0f}')\n\n# Visualization\nprint('\\n=== Visualization ===')\n\nfig, axes = plt.subplots(2, 2, figsize=(15, 12))\n\n# 1. Property predictions vs true values\nfor i, target in enumerate(['pIC50', 'logS']):\n    ax = axes[0, i]\n    perf = performance[target]\n    \n    ax.scatter(perf['true_values'], perf['predictions'], alpha=0.6, s=20)\n    \n    # Perfect prediction line\n    min_val = min(perf['true_values'].min(), perf['predictions'].min())\n    max_val = max(perf['true_values'].max(), perf['predictions'].max())\n    ax.plot([min_val, max_val], [min_val, max_val], 'r--', alpha=0.8)\n    \n    ax.set_xlabel(f'True {target}')\n    ax.set_ylabel(f'Predicted {target}')\n    ax.set_title(f'{target} Prediction (R² = {perf[\"test_r2\"]:.3f})')\n    ax.grid(True, alpha=0.3)\n\n# 2. Feature importance heatmap\nax = axes[1, 0]\nimportance_matrix = feature_importance[target_cols].values.T\nim = ax.imshow(importance_matrix, cmap='viridis', aspect='auto')\nax.set_xticks(range(len(feature_cols)))\nax.set_xticklabels(feature_cols, rotation=45, ha='right')\nax.set_yticks(range(len(target_cols)))\nax.set_yticklabels(target_cols)\nax.set_title('Feature Importance Heatmap')\nplt.colorbar(im, ax=ax)\n\n# 3. Virtual screening results\nax = axes[1, 1]\nax.scatter(screening_df['pred_pIC50'], screening_df['pred_logS'], \n          alpha=0.3, s=10, color='lightblue', label='All molecules')\nif len(hits) > 0:\n    ax.scatter(hits['pred_pIC50'], hits['pred_logS'], \n              alpha=0.8, s=30, color='red', label=f'Hits ({len(hits)})')\n\nax.axvline(x=activity_threshold, color='green', linestyle='--', alpha=0.7, \n          label=f'pIC50 > {activity_threshold}')\nax.axhline(y=solubility_threshold, color='orange', linestyle='--', alpha=0.7, \n          label=f'logS > {solubility_threshold}')\n\nax.set_xlabel('Predicted pIC50')\nax.set_ylabel('Predicted logS')\nax.set_title('Virtual Screening Results')\nax.legend()\nax.grid(True, alpha=0.3)\n\nplt.tight_layout()\n\n# Save visualization\nwith tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:\n    plt.savefig(tmp.name, dpi=150, bbox_inches='tight')\n    viz_file = tmp.name\n\nplt.close()\nprint(f'Analysis visualization saved to: {viz_file}')\n\n# Summary report\nprint('\\n' + '=' * 45)\nprint('OPENCHEM DRUG DISCOVERY SUMMARY')\nprint('=' * 45)\nprint(f'Molecules analyzed: {len(df):,}')\nprint(f'Properties predicted: {len(target_cols)}')\nprint(f'Drug-like molecules: {df[\"drug_like\"].sum():,} ({df[\"drug_like\"].mean():.1%})')\nprint(f'\\nModel performance (test R²):')\nfor target in target_cols:\n    print(f'  {target}: {performance[target][\"test_r2\"]:.3f}')\nprint(f'\\nVirtual screening:')\nprint(f'  Molecules screened: {len(screening_df):,}')\nprint(f'  Hits identified: {len(hits):,}')\nprint(f'  Hit rate: {len(hits) / len(screening_df):.2%}')\n\n# Cleanup\nos.unlink(viz_file)\nprint('\\nDemo complete - temporary files cleaned up')\n\nprint('\\nOpenChem provides:')\nprint('• Multi-task molecular property prediction')\nprint('• Graph neural networks for molecules')\nprint('• Molecular generation and optimization')\nprint('• ADMET property prediction')\nprint('• Virtual screening capabilities')\nprint('• Integration with PyTorch')\nprint('• Pre-trained models and datasets')\n\nprint('\\nTypical OpenChem workflow:')\nprint('1. Load molecular dataset (SMILES, SDF)')\nprint('2. Generate molecular representations')\nprint('3. Train/load prediction models')\nprint('4. Predict properties for new molecules')\nprint('5. Apply filters for drug discovery')",
+      "usage_example": "# OpenChem deep learning for drug discovery\n# This demonstrates molecular property prediction concepts\n\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.ensemble import RandomForestRegressor\nfrom sklearn.metrics import mean_squared_error, r2_score\nimport tempfile\nimport os\nfrom collections import defaultdict\n\nprint('OpenChem - Deep Learning for Drug Discovery')\nprint('=' * 45)\n\n# Overview of OpenChem capabilities\nprint('OpenChem Features:')\nprint('• Molecular property prediction (ADMET, bioactivity)')\nprint('• Graph neural networks for molecules')\nprint('• Molecular generation and optimization')\nprint('• Multi-task learning for drug discovery')\nprint('• Molecular descriptors and fingerprints')\nprint('• Integration with PyTorch and other ML frameworks')\n\nprint('\\nSupported Molecular Representations:')\nprint('• SMILES strings')\nprint('• Molecular graphs')\nprint('• 3D conformations')\nprint('• Molecular fingerprints (ECFP, MACCS, etc.)')\nprint('• Physicochemical descriptors')\n\n# Simulate molecular dataset\nprint('\\n=== Molecular Dataset Simulation ===')\n\nnp.random.seed(42)\n\n# Generate synthetic molecular data\nn_molecules = 1000\nprint(f'Generating {n_molecules} synthetic molecules...')\n\n# Molecular descriptors (simplified)\ndescriptor_names = [\n    'molecular_weight', 'logP', 'num_donors', 'num_acceptors',\n    'tpsa', 'num_rotatable_bonds', 'num_aromatic_rings',\n    'formal_charge', 'num_heteroatoms', 'fraction_csp3'\n]\n\n# Generate realistic molecular descriptors\nmolecular_data = []\nfor i in range(n_molecules):\n    # Molecular weight: 150-600 Da (drug-like range)\n    mw = np.random.normal(350, 100)\n    mw = np.clip(mw, 150, 600)\n    \n    # LogP: -2 to 6 (lipophilicity)\n    logP = np.random.normal(2.5, 1.5)\n    logP = np.clip(logP, -2, 6)\n    \n    # Hydrogen bond donors: 0-5\n    donors = np.random.poisson(2)\n    donors = np.clip(donors, 0, 5)\n    \n    # Hydrogen bond acceptors: 0-10\n    acceptors = np.random.poisson(4)\n    acceptors = np.clip(acceptors, 0, 10)\n    \n    # Topological polar surface area: 0-200 Ų\n    tpsa = np.random.gamma(2, 30)\n    tpsa = np.clip(tpsa, 0, 200)\n    \n    # Rotatable bonds: 0-15\n    rot_bonds = np.random.poisson(5)\n    rot_bonds = np.clip(rot_bonds, 0, 15)\n    \n    # Aromatic rings: 0-4\n    aromatic_rings = np.random.poisson(2)\n    aromatic_rings = np.clip(aromatic_rings, 0, 4)\n    \n    # Formal charge: typically 0, sometimes ±1\n    formal_charge = np.random.choice([0, 0, 0, 0, 1, -1])\n    \n    # Heteroatoms: 1-8\n    heteroatoms = np.random.poisson(3) + 1\n    heteroatoms = np.clip(heteroatoms, 1, 8)\n    \n    # Fraction sp3 carbons: 0-1\n    frac_csp3 = np.random.beta(2, 2)\n    \n    molecule = {\n        'id': f'MOL_{i:04d}',\n        'molecular_weight': mw,\n        'logP': logP,\n        'num_donors': donors,\n        'num_acceptors': acceptors,\n        'tpsa': tpsa,\n        'num_rotatable_bonds': rot_bonds,\n        'num_aromatic_rings': aromatic_rings,\n        'formal_charge': formal_charge,\n        'num_heteroatoms': heteroatoms,\n        'fraction_csp3': frac_csp3\n    }\n    \n    molecular_data.append(molecule)\n\n# Convert to DataFrame\ndf = pd.DataFrame(molecular_data)\nprint(f'Generated molecular dataset: {df.shape}')\nprint(f'Descriptors: {len(descriptor_names)}')\n\n# Show basic statistics\nprint('\\nDataset statistics:')\nprint(df[descriptor_names].describe().round(2))\n\n# Generate target properties\nprint('\\n=== Target Property Generation ===')\n\n# Simulate bioactivity (IC50 values)\nprint('Generating bioactivity data (IC50 values)...')\n\ndef calculate_bioactivity(row):\n    \"\"\"Simulate bioactivity based on molecular descriptors\"\"\"\n    # Lipinski's rule of five compliance\n    lipinski_score = 0\n    if row['molecular_weight'] <= 500: lipinski_score += 1\n    if row['logP'] <= 5: lipinski_score += 1\n    if row['num_donors'] <= 5: lipinski_score += 1\n    if row['num_acceptors'] <= 10: lipinski_score += 1\n    \n    # Base activity influenced by Lipinski compliance\n    base_activity = 5.0 + (lipinski_score - 2.0) * 1.5\n    \n    # Additional molecular factors\n    tpsa_factor = -0.01 * max(0, row['tpsa'] - 90)  # Penalty for high TPSA\n    flexibility_factor = -0.1 * max(0, row['num_rotatable_bonds'] - 7)  # Penalty for high flexibility\n    aromatic_factor = 0.3 * min(row['num_aromatic_rings'], 3)  # Bonus for aromatics (up to 3)\n    \n    # Combined activity (pIC50)\n    activity = base_activity + tpsa_factor + flexibility_factor + aromatic_factor\n    \n    # Add some noise\n    activity += np.random.normal(0, 0.5)\n    \n    # Ensure reasonable range (4-9 pIC50)\n    activity = np.clip(activity, 4.0, 9.0)\n    \n    return activity\n\n# Calculate bioactivity\ndf['pIC50'] = df.apply(calculate_bioactivity, axis=1)\n\n# Generate additional properties\nprint('Generating ADMET properties...')\n\n# Solubility (LogS)\ndef calculate_solubility(row):\n    \"\"\"Simulate aqueous solubility\"\"\"\n    # Lipophilicity penalty\n    logP_penalty = -0.5 * max(0, row['logP'] - 2)\n    \n    # Molecular weight penalty\n    mw_penalty = -0.005 * max(0, row['molecular_weight'] - 300)\n    \n    # TPSA bonus (polar surface area helps solubility)\n    tpsa_bonus = 0.01 * min(row['tpsa'], 100)\n    \n    # Base solubility\n    base_solubility = -2.0\n    \n    solubility = base_solubility + logP_penalty + mw_penalty + tpsa_bonus\n    solubility += np.random.normal(0, 0.3)\n    \n    return np.clip(solubility, -6.0, 1.0)\n\ndf['logS'] = df.apply(calculate_solubility, axis=1)\n\n# Permeability (Caco-2)\ndef calculate_permeability(row):\n    \"\"\"Simulate cell permeability\"\"\"\n    # LogP correlation\n    logP_factor = 0.3 * row['logP']\n    \n    # TPSA penalty\n    tpsa_penalty = -0.02 * row['tpsa']\n    \n    # Molecular weight penalty\n    mw_penalty = -0.003 * row['molecular_weight']\n    \n    # Base permeability (log Papp)\n    base_perm = -4.5\n    \n    permeability = base_perm + logP_factor + tpsa_penalty + mw_penalty\n    permeability += np.random.normal(0, 0.4)\n    \n    return np.clip(permeability, -7.0, -3.0)\n\ndf['log_Papp'] = df.apply(calculate_permeability, axis=1)\n\n# Half-life (t1/2)\ndef calculate_half_life(row):\n    \"\"\"Simulate metabolic stability\"\"\"\n    # Molecular complexity factor\n    complexity = row['num_rotatable_bonds'] + row['num_heteroatoms']\n    complexity_factor = -0.05 * complexity\n    \n    # Aromatic stabilization\n    aromatic_factor = 0.1 * row['num_aromatic_rings']\n    \n    # Base half-life (log hours)\n    base_t_half = 1.0\n    \n    t_half = base_t_half + complexity_factor + aromatic_factor\n    t_half += np.random.normal(0, 0.3)\n    \n    return np.clip(t_half, -1.0, 3.0)\n\ndf['log_t_half'] = df.apply(calculate_half_life, axis=1)\n\nprint(f'Generated properties: pIC50, logS, log_Papp, log_t_half')\nprint(f'Property value ranges:')\nfor prop in ['pIC50', 'logS', 'log_Papp', 'log_t_half']:\n    print(f'  {prop}: {df[prop].min():.2f} to {df[prop].max():.2f}')\n\n# Apply drug-likeness filters\nprint('\\n=== Drug-likeness Analysis ===')\n\ndef lipinski_filter(row):\n    \"\"\"Apply Lipinski's Rule of Five\"\"\"\n    violations = 0\n    if row['molecular_weight'] > 500: violations += 1\n    if row['logP'] > 5: violations += 1\n    if row['num_donors'] > 5: violations += 1\n    if row['num_acceptors'] > 10: violations += 1\n    return violations\n\ndef veber_filter(row):\n    \"\"\"Apply Veber's rules for oral bioavailability\"\"\"\n    violations = 0\n    if row['tpsa'] > 140: violations += 1\n    if row['num_rotatable_bonds'] > 10: violations += 1\n    return violations\n\ndf['lipinski_violations'] = df.apply(lipinski_filter, axis=1)\ndf['veber_violations'] = df.apply(veber_filter, axis=1)\ndf['drug_like'] = (df['lipinski_violations'] == 0) & (df['veber_violations'] == 0)\n\nprint(f'Drug-likeness analysis:')\nprint(f'  Lipinski compliant (0 violations): {(df[\"lipinski_violations\"] == 0).sum()}')\nprint(f'  Veber compliant (0 violations): {(df[\"veber_violations\"] == 0).sum()}')\nprint(f'  Overall drug-like: {df[\"drug_like\"].sum()}')\nprint(f'  Drug-like percentage: {df[\"drug_like\"].mean():.1%}')\n\n# Machine learning model training\nprint('\\n=== Machine Learning Model Training ===')\n\n# Prepare features and targets\nfeature_cols = descriptor_names\ntarget_cols = ['pIC50', 'logS', 'log_Papp', 'log_t_half']\n\nX = df[feature_cols].values\ny = df[target_cols].values\n\nprint(f'Feature matrix shape: {X.shape}')\nprint(f'Target matrix shape: {y.shape}')\n\n# Split data\nX_train, X_test, y_train, y_test = train_test_split(\n    X, y, test_size=0.2, random_state=42\n)\n\nprint(f'Training set: {X_train.shape[0]} molecules')\nprint(f'Test set: {X_test.shape[0]} molecules')\n\n# Train models for each property\nmodels = {}\nperformance = {}\n\nfor i, target_name in enumerate(target_cols):\n    print(f'\\nTraining model for {target_name}...')\n    \n    # Random Forest model\n    model = RandomForestRegressor(n_estimators=100, random_state=42)\n    model.fit(X_train, y_train[:, i])\n    \n    # Predictions\n    y_pred_train = model.predict(X_train)\n    y_pred_test = model.predict(X_test)\n    \n    # Performance metrics\n    train_r2 = r2_score(y_train[:, i], y_pred_train)\n    test_r2 = r2_score(y_test[:, i], y_pred_test)\n    train_rmse = np.sqrt(mean_squared_error(y_train[:, i], y_pred_train))\n    test_rmse = np.sqrt(mean_squared_error(y_test[:, i], y_pred_test))\n    \n    models[target_name] = model\n    performance[target_name] = {\n        'train_r2': train_r2,\n        'test_r2': test_r2,\n        'train_rmse': train_rmse,\n        'test_rmse': test_rmse,\n        'predictions': y_pred_test,\n        'true_values': y_test[:, i]\n    }\n    \n    print(f'  Training R²: {train_r2:.3f}')\n    print(f'  Test R²: {test_r2:.3f}')\n    print(f'  Test RMSE: {test_rmse:.3f}')\n\n# Feature importance analysis\nprint('\\n=== Feature Importance Analysis ===')\n\nfeature_importance = pd.DataFrame({\n    'feature': feature_cols,\n    **{target: models[target].feature_importances_ for target in target_cols}\n})\n\nprint('Top features for each property:')\nfor target in target_cols:\n    top_features = feature_importance.nlargest(3, target)\n    print(f'\\n{target}:')\n    for _, row in top_features.iterrows():\n        print(f'  {row[\"feature\"]}: {row[target]:.3f}')\n\n# Virtual screening simulation\nprint('\\n=== Virtual Screening Simulation ===')\n\n# Generate new molecules for screening\nn_screening = 10000\nprint(f'Generating {n_screening} molecules for virtual screening...')\n\nscreening_data = []\nfor i in range(n_screening):\n    # Generate random molecular descriptors\n    mw = np.random.normal(350, 120)\n    mw = np.clip(mw, 100, 700)\n    \n    logP = np.random.normal(2.5, 2.0)\n    logP = np.clip(logP, -3, 7)\n    \n    donors = np.random.poisson(2)\n    donors = np.clip(donors, 0, 8)\n    \n    acceptors = np.random.poisson(4)\n    acceptors = np.clip(acceptors, 0, 15)\n    \n    tpsa = np.random.gamma(2, 35)\n    tpsa = np.clip(tpsa, 0, 250)\n    \n    rot_bonds = np.random.poisson(6)\n    rot_bonds = np.clip(rot_bonds, 0, 20)\n    \n    aromatic_rings = np.random.poisson(2)\n    aromatic_rings = np.clip(aromatic_rings, 0, 5)\n    \n    formal_charge = np.random.choice([0, 0, 0, 0, 1, -1])\n    \n    heteroatoms = np.random.poisson(4) + 1\n    heteroatoms = np.clip(heteroatoms, 1, 12)\n    \n    frac_csp3 = np.random.beta(2, 2)\n    \n    screening_data.append([\n        mw, logP, donors, acceptors, tpsa, rot_bonds,\n        aromatic_rings, formal_charge, heteroatoms, frac_csp3\n    ])\n\nX_screening = np.array(screening_data)\n\n# Predict properties for screening library\nprint('Predicting properties for screening library...')\nscreening_predictions = {}\n\nfor target in target_cols:\n    predictions = models[target].predict(X_screening)\n    screening_predictions[target] = predictions\n\n# Apply filters for hit identification\nprint('\\nApplying screening filters...')\n\n# Create screening DataFrame\nscreening_df = pd.DataFrame(X_screening, columns=feature_cols)\nfor target in target_cols:\n    screening_df[f'pred_{target}'] = screening_predictions[target]\n\n# Apply drug-likeness filters\nscreening_df['lipinski_violations'] = screening_df.apply(lipinski_filter, axis=1)\nscreening_df['veber_violations'] = screening_df.apply(veber_filter, axis=1)\nscreening_df['drug_like'] = (\n    (screening_df['lipinski_violations'] == 0) & \n    (screening_df['veber_violations'] == 0)\n)\n\n# Activity filters\nactivity_threshold = 6.5  # pIC50 > 6.5 (IC50 < 316 nM)\nsolubility_threshold = -4.0  # logS > -4 (> 0.1 mM)\npermeability_threshold = -5.5  # log Papp > -5.5\n\nhits = screening_df[\n    (screening_df['pred_pIC50'] > activity_threshold) &\n    (screening_df['pred_logS'] > solubility_threshold) &\n    (screening_df['pred_log_Papp'] > permeability_threshold) &\n    (screening_df['drug_like'] == True)\n]\n\nprint(f'Screening results:')\nprint(f'  Total molecules: {len(screening_df):}')\nprint(f'  Drug-like molecules: {screening_df[\"drug_like\"].sum():}')\nprint(f'  Active hits (pIC50 > {activity_threshold}): {(screening_df[\"pred_pIC50\"] > activity_threshold).sum():}')\nprint(f'  Final hits (all criteria): {len(hits):}')\nprint(f'  Hit rate: {len(hits) / len(screening_df):.1%}')\n\nif len(hits) > 0:\n    print(f'\\nTop 5 hits:')\n    top_hits = hits.nlargest(5, 'pred_pIC50')\n    for idx, hit in top_hits.iterrows():\n        print(f'  Hit {idx}: pIC50={hit[\"pred_pIC50\"]:.2f}, '\n              f'logS={hit[\"pred_logS\"]:.2f}, MW={hit[\"molecular_weight\"]:.0f}')\n\n# Visualization\nprint('\\n=== Visualization ===')\n\nfig, axes = plt.subplots(2, 2, figsize=(15, 12))\n\n# 1. Property predictions vs true values\nfor i, target in enumerate(['pIC50', 'logS']):\n    ax = axes[0, i]\n    perf = performance[target]\n    \n    ax.scatter(perf['true_values'], perf['predictions'], alpha=0.6, s=20)\n    \n    # Perfect prediction line\n    min_val = min(perf['true_values'].min(), perf['predictions'].min())\n    max_val = max(perf['true_values'].max(), perf['predictions'].max())\n    ax.plot([min_val, max_val], [min_val, max_val], 'r--', alpha=0.8)\n    \n    ax.set_xlabel(f'True {target}')\n    ax.set_ylabel(f'Predicted {target}')\n    ax.set_title(f'{target} Prediction (R² = {perf[\"test_r2\"]:.3f})')\n    ax.grid(True, alpha=0.3)\n\n# 2. Feature importance heatmap\nax = axes[1, 0]\nimportance_matrix = feature_importance[target_cols].values.T\nim = ax.imshow(importance_matrix, cmap='viridis', aspect='auto')\nax.set_xticks(range(len(feature_cols)))\nax.set_xticklabels(feature_cols, rotation=45, ha='right')\nax.set_yticks(range(len(target_cols)))\nax.set_yticklabels(target_cols)\nax.set_title('Feature Importance Heatmap')\nplt.colorbar(im, ax=ax)\n\n# 3. Virtual screening results\nax = axes[1, 1]\nax.scatter(screening_df['pred_pIC50'], screening_df['pred_logS'], \n          alpha=0.3, s=10, color='lightblue', label='All molecules')\nif len(hits) > 0:\n    ax.scatter(hits['pred_pIC50'], hits['pred_logS'], \n              alpha=0.8, s=30, color='red', label=f'Hits ({len(hits)})')\n\nax.axvline(x=activity_threshold, color='green', linestyle='--', alpha=0.7, \n          label=f'pIC50 > {activity_threshold}')\nax.axhline(y=solubility_threshold, color='orange', linestyle='--', alpha=0.7, \n          label=f'logS > {solubility_threshold}')\n\nax.set_xlabel('Predicted pIC50')\nax.set_ylabel('Predicted logS')\nax.set_title('Virtual Screening Results')\nax.legend()\nax.grid(True, alpha=0.3)\n\nplt.tight_layout()\n\n# Save visualization\nwith tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:\n    plt.savefig(tmp.name, dpi=150, bbox_inches='tight')\n    viz_file = tmp.name\n\nplt.close()\nprint(f'Analysis visualization saved to: {viz_file}')\n\n# Summary report\nprint('\\n' + '=' * 45)\nprint('OPENCHEM DRUG DISCOVERY SUMMARY')\nprint('=' * 45)\nprint(f'Molecules analyzed: {len(df):}')\nprint(f'Properties predicted: {len(target_cols)}')\nprint(f'Drug-like molecules: {df[\"drug_like\"].sum():} ({df[\"drug_like\"].mean():.1%})')\nprint(f'\\nModel performance (test R²):')\nfor target in target_cols:\n    print(f'  {target}: {performance[target][\"test_r2\"]:.3f}')\nprint(f'\\nVirtual screening:')\nprint(f'  Molecules screened: {len(screening_df):}')\nprint(f'  Hits identified: {len(hits):}')\nprint(f'  Hit rate: {len(hits) / len(screening_df):.2%}')\n\n# Cleanup\nos.unlink(viz_file)\nprint('\\nDemo complete - temporary files cleaned up')\n\nprint('\\nOpenChem provides:')\nprint('• Multi-task molecular property prediction')\nprint('• Graph neural networks for molecules')\nprint('• Molecular generation and optimization')\nprint('• ADMET property prediction')\nprint('• Virtual screening capabilities')\nprint('• Integration with PyTorch')\nprint('• Pre-trained models and datasets')\n\nprint('\\nTypical OpenChem workflow:')\nprint('1. Load molecular dataset (SMILES, SDF)')\nprint('2. Generate molecular representations')\nprint('3. Train/load prediction models')\nprint('4. Predict properties for new molecules')\nprint('5. Apply filters for drug discovery')",
       "quick_start": [
         "Install: pip install openchem",
         "Import: from openchem.models import build_model",

tooluniverse/data/packages/genomics_tools.json

@@ -582,7 +582,7 @@
         "pip": "pip install pydeseq2",
         "conda": "conda install -c conda-forge pydeseq2"
       },
-      "usage_example": "import pandas as pd\nimport numpy as np\nfrom pydeseq2 import DeseqDataSet\nfrom pydeseq2.dds import DeseqStats\nfrom pydeseq2.default_inference import DefaultInference\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\nprint('PyDESeq2 - RNA-seq Differential Expression Analysis')\nprint('=' * 55)\n\n# Create synthetic RNA-seq count data\nprint('Creating synthetic RNA-seq count data...')\n\n# Set random seed for reproducibility\nnp.random.seed(42)\n\n# Parameters\nn_genes = 1000\nn_samples_per_condition = 6\nconditions = ['control', 'treatment']\ntotal_samples = n_samples_per_condition * len(conditions)\n\n# Generate gene names\ngene_names = [f'Gene_{i:04d}' for i in range(1, n_genes + 1)]\n\n# Generate sample names and metadata\nsample_names = []\ncondition_labels = []\nfor condition in conditions:\n    for i in range(n_samples_per_condition):\n        sample_names.append(f'{condition}_rep{i+1}')\n        condition_labels.append(condition)\n\n# Create metadata DataFrame\nmetadata = pd.DataFrame({\n    'sample_id': sample_names,\n    'condition': condition_labels\n})\nmetadata.set_index('sample_id', inplace=True)\n\nprint(f'Created metadata for {len(sample_names)} samples:')\nprint(metadata.groupby('condition').size())\n\n# Generate count matrix\nprint('\\nGenerating count matrix...')\n\n# Base expression levels (log scale)\nbase_expression = np.random.negative_binomial(n=5, p=0.3, size=n_genes)\n\n# Create count matrix\ncounts = np.zeros((n_genes, total_samples))\n\n# Generate counts for each sample\nfor i, condition in enumerate(condition_labels):\n    # Add some noise and condition-specific effects\n    if condition == 'control':\n        # Control condition - use base expression\n        sample_counts = np.random.negative_binomial(\n            n=base_expression, \n            p=0.1,  # Dispersion parameter\n            size=n_genes\n        )\n    else:\n        # Treatment condition - add differential expression\n        # Select ~10% of genes to be differentially expressed\n        de_genes = np.random.choice(n_genes, size=int(0.1 * n_genes), replace=False)\n        \n        modified_expression = base_expression.copy()\n        \n        # Half upregulated, half downregulated\n        up_genes = de_genes[:len(de_genes)//2]\n        down_genes = de_genes[len(de_genes)//2:]\n        \n        # Upregulate (2-4 fold)\n        modified_expression[up_genes] *= np.random.uniform(2, 4, len(up_genes))\n        \n        # Downregulate (0.25-0.5 fold)\n        modified_expression[down_genes] *= np.random.uniform(0.25, 0.5, len(down_genes))\n        \n        sample_counts = np.random.negative_binomial(\n            n=modified_expression.astype(int), \n            p=0.1,\n            size=n_genes\n        )\n    \n    counts[:, i] = sample_counts\n\n# Create count DataFrame\ncount_df = pd.DataFrame(counts, index=gene_names, columns=sample_names)\ncount_df = count_df.astype(int)\n\nprint(f'Count matrix shape: {count_df.shape}')\nprint(f'Total reads per sample:')\nfor sample in count_df.columns:\n    total_reads = count_df[sample].sum()\n    print(f'  {sample}: {total_reads:,} reads')\n\nprint(f'\\nCount statistics:')\nprint(f'  Mean counts per gene: {count_df.mean(axis=1).mean():.1f}')\nprint(f'  Median counts per gene: {count_df.median(axis=1).median():.1f}')\nprint(f'  Genes with zero counts: {(count_df.sum(axis=1) == 0).sum()}')\n\n# Filter low-count genes\nprint('\\nFiltering low-count genes...')\nmin_count = 10\nmin_samples = 3\n\n# Keep genes with at least min_count reads in at least min_samples samples\nkeep_genes = (count_df >= min_count).sum(axis=1) >= min_samples\nfiltered_counts = count_df[keep_genes]\n\nprint(f'Genes before filtering: {len(count_df)}')\nprint(f'Genes after filtering: {len(filtered_counts)}')\nprint(f'Genes removed: {len(count_df) - len(filtered_counts)}')\n\n# Create DESeq2 dataset\nprint('\\n=== DESeq2 Analysis ===')\nprint('Creating DESeq2 dataset...')\n\n# Prepare data for PyDESeq2\ninference = DefaultInference(n_cpus=1)\n\n# Create DESeq2 dataset\ndds = DeseqDataSet(\n    counts=filtered_counts,\n    metadata=metadata,\n    design_factors=['condition'],\n    refit_cooks=True,\n    inference=inference\n)\n\nprint(f'DESeq2 dataset created with {dds.n_obs} genes and {dds.n_vars} samples')\n\n# Run DESeq2 analysis\nprint('\\nRunning DESeq2 analysis...')\nprint('1. Estimating size factors...')\ndds.fit_size_factors()\n\nprint('2. Estimating dispersions...')\ndds.fit_genewise_dispersions()\ndds.fit_dispersion_trend()\ndds.fit_dispersion_prior()\ndds.fit_MAP_dispersions()\n\nprint('3. Fitting generalized linear model...')\ndds.fit_LFC()\n\nprint('4. Running statistical tests...')\nstat_res = DeseqStats(dds, inference=inference)\nstat_res.summary()\n\n# Get results\nprint('\\n=== Results Analysis ===')\nresults_df = stat_res.results_df\n\nprint(f'Results shape: {results_df.shape}')\nprint(f'Columns: {list(results_df.columns)}')\n\n# Filter for significant genes\nalpha = 0.05\nlog2fc_threshold = 1.0\n\nsignificant = (\n    (results_df['padj'] < alpha) & \n    (np.abs(results_df['log2FoldChange']) > log2fc_threshold)\n)\n\nupregulated = (\n    (results_df['padj'] < alpha) & \n    (results_df['log2FoldChange'] > log2fc_threshold)\n)\n\ndownregulated = (\n    (results_df['padj'] < alpha) & \n    (results_df['log2FoldChange'] < -log2fc_threshold)\n)\n\nprint(f'\\nDifferential expression results:')\nprint(f'  Total genes tested: {len(results_df)}')\nprint(f'  Significant genes (padj < {alpha}, |log2FC| > {log2fc_threshold}): {significant.sum()}')\nprint(f'  Upregulated genes: {upregulated.sum()}')\nprint(f'  Downregulated genes: {downregulated.sum()}')\n\n# Show top differentially expressed genes\nprint('\\nTop 10 upregulated genes:')\ntop_up = results_df[upregulated].nlargest(10, 'log2FoldChange')\nfor gene, row in top_up.iterrows():\n    print(f'  {gene}: log2FC={row[\"log2FoldChange\"]:.2f}, padj={row[\"padj\"]:.2e}')\n\nprint('\\nTop 10 downregulated genes:')\ntop_down = results_df[downregulated].nsmallest(10, 'log2FoldChange')\nfor gene, row in top_down.iterrows():\n    print(f'  {gene}: log2FC={row[\"log2FoldChange\"]:.2f}, padj={row[\"padj\"]:.2e}')\n\n# Quality control plots\nprint('\\n=== Quality Control Plots ===')\n\nfig, axes = plt.subplots(2, 2, figsize=(12, 10))\n\n# 1. MA plot\naxes[0, 0].scatter(results_df['baseMean'], results_df['log2FoldChange'], \n                  alpha=0.5, s=1, color='gray')\naxes[0, 0].scatter(results_df.loc[significant, 'baseMean'], \n                  results_df.loc[significant, 'log2FoldChange'], \n                  alpha=0.7, s=2, color='red')\naxes[0, 0].axhline(y=0, color='blue', linestyle='--', alpha=0.7)\naxes[0, 0].axhline(y=log2fc_threshold, color='green', linestyle='--', alpha=0.7)\naxes[0, 0].axhline(y=-log2fc_threshold, color='green', linestyle='--', alpha=0.7)\naxes[0, 0].set_xlabel('Mean Expression')\naxes[0, 0].set_ylabel('Log2 Fold Change')\naxes[0, 0].set_title('MA Plot')\naxes[0, 0].set_xscale('log')\n\n# 2. Volcano plot\np_values = -np.log10(results_df['padj'].fillna(1))\naxes[0, 1].scatter(results_df['log2FoldChange'], p_values, \n                  alpha=0.5, s=1, color='gray')\naxes[0, 1].scatter(results_df.loc[significant, 'log2FoldChange'], \n                  p_values[significant], \n                  alpha=0.7, s=2, color='red')\naxes[0, 1].axhline(y=-np.log10(alpha), color='green', linestyle='--', alpha=0.7)\naxes[0, 1].axvline(x=log2fc_threshold, color='green', linestyle='--', alpha=0.7)\naxes[0, 1].axvline(x=-log2fc_threshold, color='green', linestyle='--', alpha=0.7)\naxes[0, 1].set_xlabel('Log2 Fold Change')\naxes[0, 1].set_ylabel('-Log10 Adjusted P-value')\naxes[0, 1].set_title('Volcano Plot')\n\n# 3. P-value histogram\naxes[1, 0].hist(results_df['pvalue'].dropna(), bins=50, alpha=0.7, color='skyblue')\naxes[1, 0].set_xlabel('P-value')\naxes[1, 0].set_ylabel('Frequency')\naxes[1, 0].set_title('P-value Distribution')\n\n# 4. Dispersion plot\naxes[1, 1].scatter(dds.layers['normed_counts'].mean(axis=1), \n                  dds.varm['dispersions'], \n                  alpha=0.5, s=1, color='gray')\naxes[1, 1].set_xlabel('Mean Normalized Counts')\naxes[1, 1].set_ylabel('Dispersion')\naxes[1, 1].set_title('Dispersion Estimates')\naxes[1, 1].set_xscale('log')\naxes[1, 1].set_yscale('log')\n\nplt.tight_layout()\n\n# Save plots\nimport tempfile\nwith tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:\n    plt.savefig(tmp.name, dpi=150, bbox_inches='tight')\n    plot_file = tmp.name\n\nplt.close()\nprint(f'QC plots saved to: {plot_file}')\n\n# Summary statistics\nprint('\\n' + '=' * 55)\nprint('DIFFERENTIAL EXPRESSION ANALYSIS SUMMARY')\nprint('=' * 55)\nprint(f'Total genes analyzed: {len(results_df):,}')\nprint(f'Significant DE genes: {significant.sum():,} ({significant.sum()/len(results_df)*100:.1f}%)')\nprint(f'Upregulated genes: {upregulated.sum():,}')\nprint(f'Downregulated genes: {downregulated.sum():,}')\nprint(f'Significance threshold: padj < {alpha}')\nprint(f'Fold change threshold: |log2FC| > {log2fc_threshold}')\n\n# Effect size distribution\nif significant.sum() > 0:\n    sig_lfc = results_df.loc[significant, 'log2FoldChange']\n    print(f'\\nEffect size statistics (significant genes):')\n    print(f'  Mean |log2FC|: {np.abs(sig_lfc).mean():.2f}')\n    print(f'  Max upregulation: {sig_lfc.max():.2f} log2FC')\n    print(f'  Max downregulation: {sig_lfc.min():.2f} log2FC')\n\n# Cleanup\nimport os\nos.unlink(plot_file)\nprint('\\nDemo complete - temporary files cleaned up')\n\nprint('\\nPyDESeq2 provides:')\nprint('• Python implementation of DESeq2')\nprint('• Differential expression analysis')\nprint('• Size factor normalization')\nprint('• Dispersion estimation')\nprint('• Statistical testing with multiple correction')\nprint('• Integration with pandas and numpy')\nprint('• Visualization and quality control')",
+      "usage_example": "import pandas as pd\nimport numpy as np\nfrom pydeseq2 import DeseqDataSet\nfrom pydeseq2.dds import DeseqStats\nfrom pydeseq2.default_inference import DefaultInference\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\nprint('PyDESeq2 - RNA-seq Differential Expression Analysis')\nprint('=' * 55)\n\n# Create synthetic RNA-seq count data\nprint('Creating synthetic RNA-seq count data...')\n\n# Set random seed for reproducibility\nnp.random.seed(42)\n\n# Parameters\nn_genes = 1000\nn_samples_per_condition = 6\nconditions = ['control', 'treatment']\ntotal_samples = n_samples_per_condition * len(conditions)\n\n# Generate gene names\ngene_names = [f'Gene_{i:04d}' for i in range(1, n_genes + 1)]\n\n# Generate sample names and metadata\nsample_names = []\ncondition_labels = []\nfor condition in conditions:\n    for i in range(n_samples_per_condition):\n        sample_names.append(f'{condition}_rep{i+1}')\n        condition_labels.append(condition)\n\n# Create metadata DataFrame\nmetadata = pd.DataFrame({\n    'sample_id': sample_names,\n    'condition': condition_labels\n})\nmetadata.set_index('sample_id', inplace=True)\n\nprint(f'Created metadata for {len(sample_names)} samples:')\nprint(metadata.groupby('condition').size())\n\n# Generate count matrix\nprint('\\nGenerating count matrix...')\n\n# Base expression levels (log scale)\nbase_expression = np.random.negative_binomial(n=5, p=0.3, size=n_genes)\n\n# Create count matrix\ncounts = np.zeros((n_genes, total_samples))\n\n# Generate counts for each sample\nfor i, condition in enumerate(condition_labels):\n    # Add some noise and condition-specific effects\n    if condition == 'control':\n        # Control condition - use base expression\n        sample_counts = np.random.negative_binomial(\n            n=base_expression, \n            p=0.1,  # Dispersion parameter\n            size=n_genes\n        )\n    else:\n        # Treatment condition - add differential expression\n        # Select ~10% of genes to be differentially expressed\n        de_genes = np.random.choice(n_genes, size=int(0.1 * n_genes), replace=False)\n        \n        modified_expression = base_expression.copy()\n        \n        # Half upregulated, half downregulated\n        up_genes = de_genes[:len(de_genes)//2]\n        down_genes = de_genes[len(de_genes)//2:]\n        \n        # Upregulate (2-4 fold)\n        modified_expression[up_genes] *= np.random.uniform(2, 4, len(up_genes))\n        \n        # Downregulate (0.25-0.5 fold)\n        modified_expression[down_genes] *= np.random.uniform(0.25, 0.5, len(down_genes))\n        \n        sample_counts = np.random.negative_binomial(\n            n=modified_expression.astype(int), \n            p=0.1,\n            size=n_genes\n        )\n    \n    counts[:, i] = sample_counts\n\n# Create count DataFrame\ncount_df = pd.DataFrame(counts, index=gene_names, columns=sample_names)\ncount_df = count_df.astype(int)\n\nprint(f'Count matrix shape: {count_df.shape}')\nprint(f'Total reads per sample:')\nfor sample in count_df.columns:\n    total_reads = count_df[sample].sum()\n    print(f'  {sample}: {total_reads:} reads')\n\nprint(f'\\nCount statistics:')\nprint(f'  Mean counts per gene: {count_df.mean(axis=1).mean():.1f}')\nprint(f'  Median counts per gene: {count_df.median(axis=1).median():.1f}')\nprint(f'  Genes with zero counts: {(count_df.sum(axis=1) == 0).sum()}')\n\n# Filter low-count genes\nprint('\\nFiltering low-count genes...')\nmin_count = 10\nmin_samples = 3\n\n# Keep genes with at least min_count reads in at least min_samples samples\nkeep_genes = (count_df >= min_count).sum(axis=1) >= min_samples\nfiltered_counts = count_df[keep_genes]\n\nprint(f'Genes before filtering: {len(count_df)}')\nprint(f'Genes after filtering: {len(filtered_counts)}')\nprint(f'Genes removed: {len(count_df) - len(filtered_counts)}')\n\n# Create DESeq2 dataset\nprint('\\n=== DESeq2 Analysis ===')\nprint('Creating DESeq2 dataset...')\n\n# Prepare data for PyDESeq2\ninference = DefaultInference(n_cpus=1)\n\n# Create DESeq2 dataset\ndds = DeseqDataSet(\n    counts=filtered_counts,\n    metadata=metadata,\n    design_factors=['condition'],\n    refit_cooks=True,\n    inference=inference\n)\n\nprint(f'DESeq2 dataset created with {dds.n_obs} genes and {dds.n_vars} samples')\n\n# Run DESeq2 analysis\nprint('\\nRunning DESeq2 analysis...')\nprint('1. Estimating size factors...')\ndds.fit_size_factors()\n\nprint('2. Estimating dispersions...')\ndds.fit_genewise_dispersions()\ndds.fit_dispersion_trend()\ndds.fit_dispersion_prior()\ndds.fit_MAP_dispersions()\n\nprint('3. Fitting generalized linear model...')\ndds.fit_LFC()\n\nprint('4. Running statistical tests...')\nstat_res = DeseqStats(dds, inference=inference)\nstat_res.summary()\n\n# Get results\nprint('\\n=== Results Analysis ===')\nresults_df = stat_res.results_df\n\nprint(f'Results shape: {results_df.shape}')\nprint(f'Columns: {list(results_df.columns)}')\n\n# Filter for significant genes\nalpha = 0.05\nlog2fc_threshold = 1.0\n\nsignificant = (\n    (results_df['padj'] < alpha) & \n    (np.abs(results_df['log2FoldChange']) > log2fc_threshold)\n)\n\nupregulated = (\n    (results_df['padj'] < alpha) & \n    (results_df['log2FoldChange'] > log2fc_threshold)\n)\n\ndownregulated = (\n    (results_df['padj'] < alpha) & \n    (results_df['log2FoldChange'] < -log2fc_threshold)\n)\n\nprint(f'\\nDifferential expression results:')\nprint(f'  Total genes tested: {len(results_df)}')\nprint(f'  Significant genes (padj < {alpha}, |log2FC| > {log2fc_threshold}): {significant.sum()}')\nprint(f'  Upregulated genes: {upregulated.sum()}')\nprint(f'  Downregulated genes: {downregulated.sum()}')\n\n# Show top differentially expressed genes\nprint('\\nTop 10 upregulated genes:')\ntop_up = results_df[upregulated].nlargest(10, 'log2FoldChange')\nfor gene, row in top_up.iterrows():\n    print(f'  {gene}: log2FC={row[\"log2FoldChange\"]:.2f}, padj={row[\"padj\"]:.2e}')\n\nprint('\\nTop 10 downregulated genes:')\ntop_down = results_df[downregulated].nsmallest(10, 'log2FoldChange')\nfor gene, row in top_down.iterrows():\n    print(f'  {gene}: log2FC={row[\"log2FoldChange\"]:.2f}, padj={row[\"padj\"]:.2e}')\n\n# Quality control plots\nprint('\\n=== Quality Control Plots ===')\n\nfig, axes = plt.subplots(2, 2, figsize=(12, 10))\n\n# 1. MA plot\naxes[0, 0].scatter(results_df['baseMean'], results_df['log2FoldChange'], \n                  alpha=0.5, s=1, color='gray')\naxes[0, 0].scatter(results_df.loc[significant, 'baseMean'], \n                  results_df.loc[significant, 'log2FoldChange'], \n                  alpha=0.7, s=2, color='red')\naxes[0, 0].axhline(y=0, color='blue', linestyle='--', alpha=0.7)\naxes[0, 0].axhline(y=log2fc_threshold, color='green', linestyle='--', alpha=0.7)\naxes[0, 0].axhline(y=-log2fc_threshold, color='green', linestyle='--', alpha=0.7)\naxes[0, 0].set_xlabel('Mean Expression')\naxes[0, 0].set_ylabel('Log2 Fold Change')\naxes[0, 0].set_title('MA Plot')\naxes[0, 0].set_xscale('log')\n\n# 2. Volcano plot\np_values = -np.log10(results_df['padj'].fillna(1))\naxes[0, 1].scatter(results_df['log2FoldChange'], p_values, \n                  alpha=0.5, s=1, color='gray')\naxes[0, 1].scatter(results_df.loc[significant, 'log2FoldChange'], \n                  p_values[significant], \n                  alpha=0.7, s=2, color='red')\naxes[0, 1].axhline(y=-np.log10(alpha), color='green', linestyle='--', alpha=0.7)\naxes[0, 1].axvline(x=log2fc_threshold, color='green', linestyle='--', alpha=0.7)\naxes[0, 1].axvline(x=-log2fc_threshold, color='green', linestyle='--', alpha=0.7)\naxes[0, 1].set_xlabel('Log2 Fold Change')\naxes[0, 1].set_ylabel('-Log10 Adjusted P-value')\naxes[0, 1].set_title('Volcano Plot')\n\n# 3. P-value histogram\naxes[1, 0].hist(results_df['pvalue'].dropna(), bins=50, alpha=0.7, color='skyblue')\naxes[1, 0].set_xlabel('P-value')\naxes[1, 0].set_ylabel('Frequency')\naxes[1, 0].set_title('P-value Distribution')\n\n# 4. Dispersion plot\naxes[1, 1].scatter(dds.layers['normed_counts'].mean(axis=1), \n                  dds.varm['dispersions'], \n                  alpha=0.5, s=1, color='gray')\naxes[1, 1].set_xlabel('Mean Normalized Counts')\naxes[1, 1].set_ylabel('Dispersion')\naxes[1, 1].set_title('Dispersion Estimates')\naxes[1, 1].set_xscale('log')\naxes[1, 1].set_yscale('log')\n\nplt.tight_layout()\n\n# Save plots\nimport tempfile\nwith tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:\n    plt.savefig(tmp.name, dpi=150, bbox_inches='tight')\n    plot_file = tmp.name\n\nplt.close()\nprint(f'QC plots saved to: {plot_file}')\n\n# Summary statistics\nprint('\\n' + '=' * 55)\nprint('DIFFERENTIAL EXPRESSION ANALYSIS SUMMARY')\nprint('=' * 55)\nprint(f'Total genes analyzed: {len(results_df):}')\nprint(f'Significant DE genes: {significant.sum():} ({significant.sum()/len(results_df)*100:.1f}%)')\nprint(f'Upregulated genes: {upregulated.sum():}')\nprint(f'Downregulated genes: {downregulated.sum():}')\nprint(f'Significance threshold: padj < {alpha}')\nprint(f'Fold change threshold: |log2FC| > {log2fc_threshold}')\n\n# Effect size distribution\nif significant.sum() > 0:\n    sig_lfc = results_df.loc[significant, 'log2FoldChange']\n    print(f'\\nEffect size statistics (significant genes):')\n    print(f'  Mean |log2FC|: {np.abs(sig_lfc).mean():.2f}')\n    print(f'  Max upregulation: {sig_lfc.max():.2f} log2FC')\n    print(f'  Max downregulation: {sig_lfc.min():.2f} log2FC')\n\n# Cleanup\nimport os\nos.unlink(plot_file)\nprint('\\nDemo complete - temporary files cleaned up')\n\nprint('\\nPyDESeq2 provides:')\nprint('• Python implementation of DESeq2')\nprint('• Differential expression analysis')\nprint('• Size factor normalization')\nprint('• Dispersion estimation')\nprint('• Statistical testing with multiple correction')\nprint('• Integration with pandas and numpy')\nprint('• Visualization and quality control')",
       "quick_start": [
         "Install: pip install pydeseq2",
         "Create dataset: dds = DeseqDataSet(counts, metadata, design_factors)",

tooluniverse/data/packages/single_cell_tools.json

@@ -276,7 +276,7 @@
         "pip": "pip install souporcell",
         "conda": "conda install -c conda-forge souporcell"
       },
-      "usage_example": "# souporcell is primarily a command-line tool\n# Here we demonstrate the concepts and analysis workflow\n\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nfrom sklearn.cluster import KMeans\nfrom sklearn.decomposition import PCA\nfrom sklearn.metrics import adjusted_rand_score\nimport tempfile\nimport os\n\nprint('souporcell - Single-cell Genotype Clustering')\nprint('=' * 45)\n\n# Overview of souporcell workflow\nprint('souporcell Workflow:')\nprint('1. Variant calling from scRNA-seq reads')\nprint('2. Genotype matrix construction')\nprint('3. Clustering cells by genotype similarity')\nprint('4. Assignment of cells to individuals')\nprint('5. Quality control and validation')\n\nprint('\\nKey Features:')\nprint('• Handles multiplexed scRNA-seq samples')\nprint('• No prior genotype information required')\nprint('• Identifies ambient RNA contamination')\nprint('• Provides cluster assignments and QC metrics')\nprint('• Compatible with 10x Genomics data')\n\n# Simulate multiplexed single-cell data\nprint('\\n=== Simulating Multiplexed scRNA-seq Data ===')\n\nnp.random.seed(42)\n\n# Simulation parameters\nn_individuals = 4\nn_cells_per_individual = 500\nn_variants = 1000\nn_genes = 2000\n\ntotal_cells = n_individuals * n_cells_per_individual\nprint(f'Simulating {total_cells} cells from {n_individuals} individuals')\nprint(f'Using {n_variants} genetic variants and {n_genes} genes')\n\n# Generate individual genotypes\nprint('\\nGenerating individual genotypes...')\nindividual_genotypes = {}\n\nfor ind_id in range(n_individuals):\n    # Each individual has different allele frequencies\n    genotype = np.random.choice([0, 1, 2], size=n_variants, p=[0.6, 0.3, 0.1])\n    individual_genotypes[f'Individual_{ind_id}'] = genotype\n\nprint(f'Generated genotypes for {len(individual_genotypes)} individuals')\n\n# Calculate genotype differences between individuals\ngenotype_matrix = np.array([geno for geno in individual_genotypes.values()])\nprint(f'Genotype matrix shape: {genotype_matrix.shape}')\n\n# Pairwise differences\nprint('\\nPairwise genotype differences:')\nfor i in range(n_individuals):\n    for j in range(i+1, n_individuals):\n        diff = np.sum(genotype_matrix[i] != genotype_matrix[j])\n        similarity = 1 - (diff / n_variants)\n        print(f'  Individual_{i} vs Individual_{j}: {diff} differences ({similarity:.3f} similarity)')\n\n# Generate cell-level data\nprint('\\nGenerating cell-level genotype data...')\n\ncell_genotypes = []\ncell_labels = []\ncell_ids = []\n\nfor ind_id in range(n_individuals):\n    individual_geno = individual_genotypes[f'Individual_{ind_id}']\n    \n    for cell_id in range(n_cells_per_individual):\n        # Add noise to simulate technical variation and allelic dropout\n        cell_geno = individual_geno.copy()\n        \n        # Simulate allelic dropout (some variants not detected)\n        dropout_rate = 0.1\n        dropout_mask = np.random.random(n_variants) < dropout_rate\n        cell_geno[dropout_mask] = 0  # Set to homozygous reference\n        \n        # Add some random noise (technical errors)\n        noise_rate = 0.02\n        noise_mask = np.random.random(n_variants) < noise_rate\n        cell_geno[noise_mask] = np.random.choice([0, 1, 2], size=np.sum(noise_mask))\n        \n        cell_genotypes.append(cell_geno)\n        cell_labels.append(ind_id)\n        cell_ids.append(f'Cell_{ind_id}_{cell_id}')\n\ncell_genotype_matrix = np.array(cell_genotypes)\nprint(f'Cell genotype matrix shape: {cell_genotype_matrix.shape}')\nprint(f'Cells per individual: {[cell_labels.count(i) for i in range(n_individuals)]}')\n\n# Add ambient RNA contamination (doublets)\nprint('\\nSimulating ambient RNA contamination (doublets)...')\nn_doublets = 100\n\nfor doublet_id in range(n_doublets):\n    # Mix genotypes from two random individuals\n    ind1, ind2 = np.random.choice(n_individuals, size=2, replace=False)\n    \n    geno1 = individual_genotypes[f'Individual_{ind1}']\n    geno2 = individual_genotypes[f'Individual_{ind2}']\n    \n    # Create mixed genotype (roughly 50:50 mix)\n    mixed_geno = np.where(np.random.random(n_variants) < 0.5, geno1, geno2)\n    \n    # Add to cell data\n    cell_genotypes.append(mixed_geno)\n    cell_labels.append(-1)  # Doublet label\n    cell_ids.append(f'Doublet_{doublet_id}')\n\n# Update matrices\ncell_genotype_matrix = np.array(cell_genotypes)\ntotal_cells_with_doublets = len(cell_genotypes)\n\nprint(f'Total cells (including doublets): {total_cells_with_doublets}')\nprint(f'Doublets added: {n_doublets}')\nprint(f'Singlets: {total_cells_with_doublets - n_doublets}')\n\n# Dimensionality reduction for visualization\nprint('\\n=== Genotype-based Clustering Analysis ===')\n\n# PCA on genotype data\nprint('Performing PCA on genotype matrix...')\npca = PCA(n_components=10)\npca_result = pca.fit_transform(cell_genotype_matrix)\n\nprint(f'PCA explained variance ratio (first 5 components): {pca.explained_variance_ratio_[:5]}')\nprint(f'Cumulative explained variance (first 5): {np.cumsum(pca.explained_variance_ratio_[:5])}')\n\n# K-means clustering\nprint('\\nPerforming K-means clustering...')\n\n# Try different numbers of clusters\ncluster_range = range(2, 8)\ninertias = []\nari_scores = []\n\nfor k in cluster_range:\n    kmeans = KMeans(n_clusters=k, random_state=42, n_init=10)\n    cluster_labels = kmeans.fit_predict(pca_result[:, :5])  # Use first 5 PCs\n    \n    inertias.append(kmeans.inertia_)\n    \n    # Calculate ARI against true labels (excluding doublets)\n    true_labels_clean = [label for label in cell_labels if label != -1]\n    cluster_labels_clean = cluster_labels[:len(true_labels_clean)]\n    \n    if len(set(true_labels_clean)) > 1 and len(set(cluster_labels_clean)) > 1:\n        ari = adjusted_rand_score(true_labels_clean, cluster_labels_clean)\n        ari_scores.append(ari)\n    else:\n        ari_scores.append(0)\n\nprint(f'Inertias for k=2 to 7: {[f\"{inertia:.0f}\" for inertia in inertias]}')\nprint(f'ARI scores for k=2 to 7: {[f\"{ari:.3f}\" for ari in ari_scores]}')\n\n# Choose optimal k (highest ARI)\nbest_k = cluster_range[np.argmax(ari_scores)]\nprint(f'\\nOptimal number of clusters: {best_k} (ARI: {max(ari_scores):.3f})')\n\n# Final clustering with optimal k\nfinal_kmeans = KMeans(n_clusters=best_k, random_state=42, n_init=10)\nfinal_clusters = final_kmeans.fit_predict(pca_result[:, :5])\n\n# Analyze cluster assignments\nprint('\\n=== Cluster Assignment Analysis ===')\n\n# Create assignment matrix\ncluster_assignment = pd.DataFrame({\n    'cell_id': cell_ids,\n    'true_individual': cell_labels,\n    'predicted_cluster': final_clusters,\n    'is_doublet': [label == -1 for label in cell_labels]\n})\n\nprint(f'Cluster assignment summary:')\nprint(cluster_assignment.groupby(['true_individual', 'predicted_cluster']).size().unstack(fill_value=0))\n\n# Calculate cluster purity\nprint('\\nCluster purity analysis:')\nfor cluster_id in range(best_k):\n    cluster_cells = cluster_assignment[cluster_assignment['predicted_cluster'] == cluster_id]\n    \n    if len(cluster_cells) > 0:\n        # Exclude doublets from purity calculation\n        singlets_in_cluster = cluster_cells[~cluster_cells['is_doublet']]\n        \n        if len(singlets_in_cluster) > 0:\n            most_common_individual = singlets_in_cluster['true_individual'].mode()\n            if len(most_common_individual) > 0:\n                purity = (singlets_in_cluster['true_individual'] == most_common_individual.iloc[0]).mean()\n                print(f'  Cluster {cluster_id}: {len(cluster_cells)} cells, '\n                      f'purity = {purity:.3f}, '\n                      f'doublets = {cluster_cells[\"is_doublet\"].sum()}')\n\n# Doublet detection analysis\nprint('\\n=== Doublet Detection Analysis ===')\n\n# Cells in clusters with mixed individuals are potential doublets\ndoublet_scores = []\n\nfor idx, row in cluster_assignment.iterrows():\n    cluster_id = row['predicted_cluster']\n    cluster_cells = cluster_assignment[cluster_assignment['predicted_cluster'] == cluster_id]\n    \n    # Calculate heterogeneity score for this cluster\n    singlets_in_cluster = cluster_cells[~cluster_cells['is_doublet']]\n    \n    if len(singlets_in_cluster) > 1:\n        individual_counts = singlets_in_cluster['true_individual'].value_counts()\n        heterogeneity = 1 - (individual_counts.max() / len(singlets_in_cluster))\n    else:\n        heterogeneity = 0\n    \n    doublet_scores.append(heterogeneity)\n\ncluster_assignment['doublet_score'] = doublet_scores\n\n# Set threshold for doublet detection\ndoublet_threshold = 0.3\npredicted_doublets = cluster_assignment['doublet_score'] > doublet_threshold\n\n# Evaluate doublet detection\ntrue_doublets = cluster_assignment['is_doublet']\ndoublet_tp = sum(predicted_doublets & true_doublets)\ndoublet_fp = sum(predicted_doublets & ~true_doublets)\ndoublet_fn = sum(~predicted_doublets & true_doublets)\ndoublet_tn = sum(~predicted_doublets & ~true_doublets)\n\ndoublet_precision = doublet_tp / (doublet_tp + doublet_fp) if (doublet_tp + doublet_fp) > 0 else 0\ndoublet_recall = doublet_tp / (doublet_tp + doublet_fn) if (doublet_tp + doublet_fn) > 0 else 0\n\nprint(f'Doublet detection performance:')\nprint(f'  True doublets: {sum(true_doublets)}')\nprint(f'  Predicted doublets: {sum(predicted_doublets)}')\nprint(f'  Precision: {doublet_precision:.3f}')\nprint(f'  Recall: {doublet_recall:.3f}')\n\n# Quality control metrics\nprint('\\n=== Quality Control Metrics ===')\n\n# Calculate per-cell variant detection rate\nvariant_detection_rates = []\nfor cell_geno in cell_genotype_matrix:\n    non_zero_variants = np.sum(cell_geno > 0)\n    detection_rate = non_zero_variants / n_variants\n    variant_detection_rates.append(detection_rate)\n\ncluster_assignment['variant_detection_rate'] = variant_detection_rates\n\nprint(f'Variant detection rates:')\nprint(f'  Mean: {np.mean(variant_detection_rates):.3f}')\nprint(f'  Median: {np.median(variant_detection_rates):.3f}')\nprint(f'  Range: {np.min(variant_detection_rates):.3f} - {np.max(variant_detection_rates):.3f}')\n\n# Per-individual statistics\nprint(f'\\nPer-individual assignment accuracy:')\nfor ind_id in range(n_individuals):\n    individual_cells = cluster_assignment[cluster_assignment['true_individual'] == ind_id]\n    \n    if len(individual_cells) > 0:\n        # Most common cluster assignment\n        most_common_cluster = individual_cells['predicted_cluster'].mode()\n        if len(most_common_cluster) > 0:\n            accuracy = (individual_cells['predicted_cluster'] == most_common_cluster.iloc[0]).mean()\n            print(f'  Individual {ind_id}: {accuracy:.3f} ({len(individual_cells)} cells)')\n\n# Visualization\nprint('\\n=== Visualization ===')\n\nfig, axes = plt.subplots(2, 2, figsize=(15, 12))\n\n# 1. PCA plot colored by true individual\nscatter1 = axes[0, 0].scatter(pca_result[:, 0], pca_result[:, 1], \n                             c=cell_labels, cmap='tab10', alpha=0.6, s=20)\naxes[0, 0].set_xlabel(f'PC1 ({pca.explained_variance_ratio_[0]:.1%} variance)')\naxes[0, 0].set_ylabel(f'PC2 ({pca.explained_variance_ratio_[1]:.1%} variance)')\naxes[0, 0].set_title('PCA - True Individuals')\n\n# 2. PCA plot colored by predicted cluster\nscatter2 = axes[0, 1].scatter(pca_result[:, 0], pca_result[:, 1], \n                             c=final_clusters, cmap='tab10', alpha=0.6, s=20)\naxes[0, 1].set_xlabel(f'PC1 ({pca.explained_variance_ratio_[0]:.1%} variance)')\naxes[0, 1].set_ylabel(f'PC2 ({pca.explained_variance_ratio_[1]:.1%} variance)')\naxes[0, 1].set_title('PCA - Predicted Clusters')\n\n# 3. Clustering evaluation metrics\nmetrics = ['Elbow Method', 'ARI Score']\naxes[1, 0].plot(cluster_range, inertias, 'bo-', label='Inertia')\naxes[1, 0].set_xlabel('Number of Clusters')\naxes[1, 0].set_ylabel('Inertia', color='blue')\naxes[1, 0].set_title('Clustering Evaluation')\naxes[1, 0].tick_params(axis='y', labelcolor='blue')\n\n# Secondary y-axis for ARI\nax_twin = axes[1, 0].twinx()\nax_twin.plot(cluster_range, ari_scores, 'ro-', label='ARI')\nax_twin.set_ylabel('Adjusted Rand Index', color='red')\nax_twin.tick_params(axis='y', labelcolor='red')\nax_twin.axvline(x=best_k, color='green', linestyle='--', alpha=0.7, label=f'Optimal k={best_k}')\n\n# 4. Doublet score distribution\naxes[1, 1].hist(cluster_assignment[cluster_assignment['is_doublet']]['doublet_score'], \n               alpha=0.7, label='True doublets', bins=20, color='red')\naxes[1, 1].hist(cluster_assignment[~cluster_assignment['is_doublet']]['doublet_score'], \n               alpha=0.7, label='Singlets', bins=20, color='blue')\naxes[1, 1].axvline(x=doublet_threshold, color='green', linestyle='--', \n                  label=f'Threshold ({doublet_threshold})')\naxes[1, 1].set_xlabel('Doublet Score')\naxes[1, 1].set_ylabel('Count')\naxes[1, 1].set_title('Doublet Score Distribution')\naxes[1, 1].legend()\n\nplt.tight_layout()\n\n# Save visualization\nwith tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:\n    plt.savefig(tmp.name, dpi=150, bbox_inches='tight')\n    viz_file = tmp.name\n\nplt.close()\nprint(f'Analysis visualization saved to: {viz_file}')\n\n# Summary report\nprint('\\n' + '=' * 45)\nprint('SOUPORCELL ANALYSIS SUMMARY')\nprint('=' * 45)\nprint(f'Total cells analyzed: {total_cells_with_doublets:,}')\nprint(f'True individuals: {n_individuals}')\nprint(f'Predicted clusters: {best_k}')\nprint(f'Clustering accuracy (ARI): {max(ari_scores):.3f}')\nprint(f'\\nDoublet detection:')\nprint(f'  True doublets: {sum(true_doublets)}')\nprint(f'  Detected doublets: {sum(predicted_doublets)}')\nprint(f'  Precision: {doublet_precision:.3f}')\nprint(f'  Recall: {doublet_recall:.3f}')\nprint(f'\\nQuality metrics:')\nprint(f'  Mean variant detection rate: {np.mean(variant_detection_rates):.3f}')\nprint(f'  Genetic variants used: {n_variants:,}')\n\n# Cleanup\nos.unlink(viz_file)\nprint('\\nDemo complete - temporary files cleaned up')\n\nprint('\\nsouporcell provides:')\nprint('• Genotype-based cell clustering')\nprint('• Multiplexed sample demultiplexing')\nprint('• Doublet detection and removal')\nprint('• Quality control metrics')\nprint('• Integration with standard scRNA-seq pipelines')\nprint('• Support for 10x Genomics data')\nprint('• Ambient RNA contamination detection')\n\nprint('\\nTypical souporcell command:')\nprint('souporcell_pipeline.py -i possorted_genome_bam.bam \\\\')\nprint('  -b filtered_feature_bc_matrix -f reference.fasta \\\\')\nprint('  -t 8 -o souporcell_output -k 4')",
+      "usage_example": "# souporcell is primarily a command-line tool\n# Here we demonstrate the concepts and analysis workflow\n\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nfrom sklearn.cluster import KMeans\nfrom sklearn.decomposition import PCA\nfrom sklearn.metrics import adjusted_rand_score\nimport tempfile\nimport os\n\nprint('souporcell - Single-cell Genotype Clustering')\nprint('=' * 45)\n\n# Overview of souporcell workflow\nprint('souporcell Workflow:')\nprint('1. Variant calling from scRNA-seq reads')\nprint('2. Genotype matrix construction')\nprint('3. Clustering cells by genotype similarity')\nprint('4. Assignment of cells to individuals')\nprint('5. Quality control and validation')\n\nprint('\\nKey Features:')\nprint('• Handles multiplexed scRNA-seq samples')\nprint('• No prior genotype information required')\nprint('• Identifies ambient RNA contamination')\nprint('• Provides cluster assignments and QC metrics')\nprint('• Compatible with 10x Genomics data')\n\n# Simulate multiplexed single-cell data\nprint('\\n=== Simulating Multiplexed scRNA-seq Data ===')\n\nnp.random.seed(42)\n\n# Simulation parameters\nn_individuals = 4\nn_cells_per_individual = 500\nn_variants = 1000\nn_genes = 2000\n\ntotal_cells = n_individuals * n_cells_per_individual\nprint(f'Simulating {total_cells} cells from {n_individuals} individuals')\nprint(f'Using {n_variants} genetic variants and {n_genes} genes')\n\n# Generate individual genotypes\nprint('\\nGenerating individual genotypes...')\nindividual_genotypes = {}\n\nfor ind_id in range(n_individuals):\n    # Each individual has different allele frequencies\n    genotype = np.random.choice([0, 1, 2], size=n_variants, p=[0.6, 0.3, 0.1])\n    individual_genotypes[f'Individual_{ind_id}'] = genotype\n\nprint(f'Generated genotypes for {len(individual_genotypes)} individuals')\n\n# Calculate genotype differences between individuals\ngenotype_matrix = np.array([geno for geno in individual_genotypes.values()])\nprint(f'Genotype matrix shape: {genotype_matrix.shape}')\n\n# Pairwise differences\nprint('\\nPairwise genotype differences:')\nfor i in range(n_individuals):\n    for j in range(i+1, n_individuals):\n        diff = np.sum(genotype_matrix[i] != genotype_matrix[j])\n        similarity = 1 - (diff / n_variants)\n        print(f'  Individual_{i} vs Individual_{j}: {diff} differences ({similarity:.3f} similarity)')\n\n# Generate cell-level data\nprint('\\nGenerating cell-level genotype data...')\n\ncell_genotypes = []\ncell_labels = []\ncell_ids = []\n\nfor ind_id in range(n_individuals):\n    individual_geno = individual_genotypes[f'Individual_{ind_id}']\n    \n    for cell_id in range(n_cells_per_individual):\n        # Add noise to simulate technical variation and allelic dropout\n        cell_geno = individual_geno.copy()\n        \n        # Simulate allelic dropout (some variants not detected)\n        dropout_rate = 0.1\n        dropout_mask = np.random.random(n_variants) < dropout_rate\n        cell_geno[dropout_mask] = 0  # Set to homozygous reference\n        \n        # Add some random noise (technical errors)\n        noise_rate = 0.02\n        noise_mask = np.random.random(n_variants) < noise_rate\n        cell_geno[noise_mask] = np.random.choice([0, 1, 2], size=np.sum(noise_mask))\n        \n        cell_genotypes.append(cell_geno)\n        cell_labels.append(ind_id)\n        cell_ids.append(f'Cell_{ind_id}_{cell_id}')\n\ncell_genotype_matrix = np.array(cell_genotypes)\nprint(f'Cell genotype matrix shape: {cell_genotype_matrix.shape}')\nprint(f'Cells per individual: {[cell_labels.count(i) for i in range(n_individuals)]}')\n\n# Add ambient RNA contamination (doublets)\nprint('\\nSimulating ambient RNA contamination (doublets)...')\nn_doublets = 100\n\nfor doublet_id in range(n_doublets):\n    # Mix genotypes from two random individuals\n    ind1, ind2 = np.random.choice(n_individuals, size=2, replace=False)\n    \n    geno1 = individual_genotypes[f'Individual_{ind1}']\n    geno2 = individual_genotypes[f'Individual_{ind2}']\n    \n    # Create mixed genotype (roughly 50:50 mix)\n    mixed_geno = np.where(np.random.random(n_variants) < 0.5, geno1, geno2)\n    \n    # Add to cell data\n    cell_genotypes.append(mixed_geno)\n    cell_labels.append(-1)  # Doublet label\n    cell_ids.append(f'Doublet_{doublet_id}')\n\n# Update matrices\ncell_genotype_matrix = np.array(cell_genotypes)\ntotal_cells_with_doublets = len(cell_genotypes)\n\nprint(f'Total cells (including doublets): {total_cells_with_doublets}')\nprint(f'Doublets added: {n_doublets}')\nprint(f'Singlets: {total_cells_with_doublets - n_doublets}')\n\n# Dimensionality reduction for visualization\nprint('\\n=== Genotype-based Clustering Analysis ===')\n\n# PCA on genotype data\nprint('Performing PCA on genotype matrix...')\npca = PCA(n_components=10)\npca_result = pca.fit_transform(cell_genotype_matrix)\n\nprint(f'PCA explained variance ratio (first 5 components): {pca.explained_variance_ratio_[:5]}')\nprint(f'Cumulative explained variance (first 5): {np.cumsum(pca.explained_variance_ratio_[:5])}')\n\n# K-means clustering\nprint('\\nPerforming K-means clustering...')\n\n# Try different numbers of clusters\ncluster_range = range(2, 8)\ninertias = []\nari_scores = []\n\nfor k in cluster_range:\n    kmeans = KMeans(n_clusters=k, random_state=42, n_init=10)\n    cluster_labels = kmeans.fit_predict(pca_result[:, :5])  # Use first 5 PCs\n    \n    inertias.append(kmeans.inertia_)\n    \n    # Calculate ARI against true labels (excluding doublets)\n    true_labels_clean = [label for label in cell_labels if label != -1]\n    cluster_labels_clean = cluster_labels[:len(true_labels_clean)]\n    \n    if len(set(true_labels_clean)) > 1 and len(set(cluster_labels_clean)) > 1:\n        ari = adjusted_rand_score(true_labels_clean, cluster_labels_clean)\n        ari_scores.append(ari)\n    else:\n        ari_scores.append(0)\n\nprint(f'Inertias for k=2 to 7: {[f\"{inertia:.0f}\" for inertia in inertias]}')\nprint(f'ARI scores for k=2 to 7: {[f\"{ari:.3f}\" for ari in ari_scores]}')\n\n# Choose optimal k (highest ARI)\nbest_k = cluster_range[np.argmax(ari_scores)]\nprint(f'\\nOptimal number of clusters: {best_k} (ARI: {max(ari_scores):.3f})')\n\n# Final clustering with optimal k\nfinal_kmeans = KMeans(n_clusters=best_k, random_state=42, n_init=10)\nfinal_clusters = final_kmeans.fit_predict(pca_result[:, :5])\n\n# Analyze cluster assignments\nprint('\\n=== Cluster Assignment Analysis ===')\n\n# Create assignment matrix\ncluster_assignment = pd.DataFrame({\n    'cell_id': cell_ids,\n    'true_individual': cell_labels,\n    'predicted_cluster': final_clusters,\n    'is_doublet': [label == -1 for label in cell_labels]\n})\n\nprint(f'Cluster assignment summary:')\nprint(cluster_assignment.groupby(['true_individual', 'predicted_cluster']).size().unstack(fill_value=0))\n\n# Calculate cluster purity\nprint('\\nCluster purity analysis:')\nfor cluster_id in range(best_k):\n    cluster_cells = cluster_assignment[cluster_assignment['predicted_cluster'] == cluster_id]\n    \n    if len(cluster_cells) > 0:\n        # Exclude doublets from purity calculation\n        singlets_in_cluster = cluster_cells[~cluster_cells['is_doublet']]\n        \n        if len(singlets_in_cluster) > 0:\n            most_common_individual = singlets_in_cluster['true_individual'].mode()\n            if len(most_common_individual) > 0:\n                purity = (singlets_in_cluster['true_individual'] == most_common_individual.iloc[0]).mean()\n                print(f'  Cluster {cluster_id}: {len(cluster_cells)} cells, '\n                      f'purity = {purity:.3f}, '\n                      f'doublets = {cluster_cells[\"is_doublet\"].sum()}')\n\n# Doublet detection analysis\nprint('\\n=== Doublet Detection Analysis ===')\n\n# Cells in clusters with mixed individuals are potential doublets\ndoublet_scores = []\n\nfor idx, row in cluster_assignment.iterrows():\n    cluster_id = row['predicted_cluster']\n    cluster_cells = cluster_assignment[cluster_assignment['predicted_cluster'] == cluster_id]\n    \n    # Calculate heterogeneity score for this cluster\n    singlets_in_cluster = cluster_cells[~cluster_cells['is_doublet']]\n    \n    if len(singlets_in_cluster) > 1:\n        individual_counts = singlets_in_cluster['true_individual'].value_counts()\n        heterogeneity = 1 - (individual_counts.max() / len(singlets_in_cluster))\n    else:\n        heterogeneity = 0\n    \n    doublet_scores.append(heterogeneity)\n\ncluster_assignment['doublet_score'] = doublet_scores\n\n# Set threshold for doublet detection\ndoublet_threshold = 0.3\npredicted_doublets = cluster_assignment['doublet_score'] > doublet_threshold\n\n# Evaluate doublet detection\ntrue_doublets = cluster_assignment['is_doublet']\ndoublet_tp = sum(predicted_doublets & true_doublets)\ndoublet_fp = sum(predicted_doublets & ~true_doublets)\ndoublet_fn = sum(~predicted_doublets & true_doublets)\ndoublet_tn = sum(~predicted_doublets & ~true_doublets)\n\ndoublet_precision = doublet_tp / (doublet_tp + doublet_fp) if (doublet_tp + doublet_fp) > 0 else 0\ndoublet_recall = doublet_tp / (doublet_tp + doublet_fn) if (doublet_tp + doublet_fn) > 0 else 0\n\nprint(f'Doublet detection performance:')\nprint(f'  True doublets: {sum(true_doublets)}')\nprint(f'  Predicted doublets: {sum(predicted_doublets)}')\nprint(f'  Precision: {doublet_precision:.3f}')\nprint(f'  Recall: {doublet_recall:.3f}')\n\n# Quality control metrics\nprint('\\n=== Quality Control Metrics ===')\n\n# Calculate per-cell variant detection rate\nvariant_detection_rates = []\nfor cell_geno in cell_genotype_matrix:\n    non_zero_variants = np.sum(cell_geno > 0)\n    detection_rate = non_zero_variants / n_variants\n    variant_detection_rates.append(detection_rate)\n\ncluster_assignment['variant_detection_rate'] = variant_detection_rates\n\nprint(f'Variant detection rates:')\nprint(f'  Mean: {np.mean(variant_detection_rates):.3f}')\nprint(f'  Median: {np.median(variant_detection_rates):.3f}')\nprint(f'  Range: {np.min(variant_detection_rates):.3f} - {np.max(variant_detection_rates):.3f}')\n\n# Per-individual statistics\nprint(f'\\nPer-individual assignment accuracy:')\nfor ind_id in range(n_individuals):\n    individual_cells = cluster_assignment[cluster_assignment['true_individual'] == ind_id]\n    \n    if len(individual_cells) > 0:\n        # Most common cluster assignment\n        most_common_cluster = individual_cells['predicted_cluster'].mode()\n        if len(most_common_cluster) > 0:\n            accuracy = (individual_cells['predicted_cluster'] == most_common_cluster.iloc[0]).mean()\n            print(f'  Individual {ind_id}: {accuracy:.3f} ({len(individual_cells)} cells)')\n\n# Visualization\nprint('\\n=== Visualization ===')\n\nfig, axes = plt.subplots(2, 2, figsize=(15, 12))\n\n# 1. PCA plot colored by true individual\nscatter1 = axes[0, 0].scatter(pca_result[:, 0], pca_result[:, 1], \n                             c=cell_labels, cmap='tab10', alpha=0.6, s=20)\naxes[0, 0].set_xlabel(f'PC1 ({pca.explained_variance_ratio_[0]:.1%} variance)')\naxes[0, 0].set_ylabel(f'PC2 ({pca.explained_variance_ratio_[1]:.1%} variance)')\naxes[0, 0].set_title('PCA - True Individuals')\n\n# 2. PCA plot colored by predicted cluster\nscatter2 = axes[0, 1].scatter(pca_result[:, 0], pca_result[:, 1], \n                             c=final_clusters, cmap='tab10', alpha=0.6, s=20)\naxes[0, 1].set_xlabel(f'PC1 ({pca.explained_variance_ratio_[0]:.1%} variance)')\naxes[0, 1].set_ylabel(f'PC2 ({pca.explained_variance_ratio_[1]:.1%} variance)')\naxes[0, 1].set_title('PCA - Predicted Clusters')\n\n# 3. Clustering evaluation metrics\nmetrics = ['Elbow Method', 'ARI Score']\naxes[1, 0].plot(cluster_range, inertias, 'bo-', label='Inertia')\naxes[1, 0].set_xlabel('Number of Clusters')\naxes[1, 0].set_ylabel('Inertia', color='blue')\naxes[1, 0].set_title('Clustering Evaluation')\naxes[1, 0].tick_params(axis='y', labelcolor='blue')\n\n# Secondary y-axis for ARI\nax_twin = axes[1, 0].twinx()\nax_twin.plot(cluster_range, ari_scores, 'ro-', label='ARI')\nax_twin.set_ylabel('Adjusted Rand Index', color='red')\nax_twin.tick_params(axis='y', labelcolor='red')\nax_twin.axvline(x=best_k, color='green', linestyle='--', alpha=0.7, label=f'Optimal k={best_k}')\n\n# 4. Doublet score distribution\naxes[1, 1].hist(cluster_assignment[cluster_assignment['is_doublet']]['doublet_score'], \n               alpha=0.7, label='True doublets', bins=20, color='red')\naxes[1, 1].hist(cluster_assignment[~cluster_assignment['is_doublet']]['doublet_score'], \n               alpha=0.7, label='Singlets', bins=20, color='blue')\naxes[1, 1].axvline(x=doublet_threshold, color='green', linestyle='--', \n                  label=f'Threshold ({doublet_threshold})')\naxes[1, 1].set_xlabel('Doublet Score')\naxes[1, 1].set_ylabel('Count')\naxes[1, 1].set_title('Doublet Score Distribution')\naxes[1, 1].legend()\n\nplt.tight_layout()\n\n# Save visualization\nwith tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:\n    plt.savefig(tmp.name, dpi=150, bbox_inches='tight')\n    viz_file = tmp.name\n\nplt.close()\nprint(f'Analysis visualization saved to: {viz_file}')\n\n# Summary report\nprint('\\n' + '=' * 45)\nprint('SOUPORCELL ANALYSIS SUMMARY')\nprint('=' * 45)\nprint(f'Total cells analyzed: {total_cells_with_doublets:}')\nprint(f'True individuals: {n_individuals}')\nprint(f'Predicted clusters: {best_k}')\nprint(f'Clustering accuracy (ARI): {max(ari_scores):.3f}')\nprint(f'\\nDoublet detection:')\nprint(f'  True doublets: {sum(true_doublets)}')\nprint(f'  Detected doublets: {sum(predicted_doublets)}')\nprint(f'  Precision: {doublet_precision:.3f}')\nprint(f'  Recall: {doublet_recall:.3f}')\nprint(f'\\nQuality metrics:')\nprint(f'  Mean variant detection rate: {np.mean(variant_detection_rates):.3f}')\nprint(f'  Genetic variants used: {n_variants:}')\n\n# Cleanup\nos.unlink(viz_file)\nprint('\\nDemo complete - temporary files cleaned up')\n\nprint('\\nsouporcell provides:')\nprint('• Genotype-based cell clustering')\nprint('• Multiplexed sample demultiplexing')\nprint('• Doublet detection and removal')\nprint('• Quality control metrics')\nprint('• Integration with standard scRNA-seq pipelines')\nprint('• Support for 10x Genomics data')\nprint('• Ambient RNA contamination detection')\n\nprint('\\nTypical souporcell command:')\nprint('souporcell_pipeline.py -i possorted_genome_bam.bam \\\\')\nprint('  -b filtered_feature_bc_matrix -f reference.fasta \\\\')\nprint('  -t 8 -o souporcell_output -k 4')",
       "quick_start": [
         "Install: pip install souporcell",
         "Run: souporcell_pipeline.py -i input.bam -b barcodes",

tooluniverse/data/packages/structural_biology_tools.json

@@ -281,7 +281,7 @@
         "pip": "pip install diffdock",
         "conda": "conda install -c conda-forge diffdock"
       },
-      "usage_example": "# DiffDock molecular docking simulation\n# This demonstrates the concepts and workflow\n\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom scipy.spatial.distance import cdist\nfrom sklearn.cluster import DBSCAN\nimport tempfile\nimport os\n\nprint('DiffDock - Diffusion Model for Molecular Docking')\nprint('=' * 50)\n\n# Overview of DiffDock approach\nprint('DiffDock Innovation:')\nprint('• Uses diffusion models for pose generation')\nprint('• End-to-end learning without hand-crafted features')\nprint('• Handles both rigid and flexible docking')\nprint('• State-of-the-art accuracy on benchmarks')\nprint('• Fast inference with GPU acceleration')\n\nprint('\\nKey Advantages over Traditional Docking:')\nprint('• No need for extensive sampling')\nprint('• Better handling of conformational flexibility')\nprint('• Learned representations from large datasets')\nprint('• Robust to protein structure variations')\n\n# Simulate molecular docking workflow\nprint('\\n=== Molecular Docking Simulation ===')\n\nnp.random.seed(42)\n\n# Define protein binding site\nprint('Defining protein binding site...')\nbinding_site_center = np.array([0.0, 0.0, 0.0])\nbinding_site_radius = 8.0\n\n# Generate protein surface points\nn_surface_points = 200\ntheta = np.random.uniform(0, 2*np.pi, n_surface_points)\nphi = np.random.uniform(0, np.pi, n_surface_points)\nr = np.random.uniform(5.0, 8.0, n_surface_points)\n\nprotein_surface = np.column_stack([\n    r * np.sin(phi) * np.cos(theta),\n    r * np.sin(phi) * np.sin(theta),\n    r * np.cos(phi)\n])\n\nprint(f'Generated {n_surface_points} protein surface points')\nprint(f'Binding site radius: {binding_site_radius:.1f} Å')\n\n# Define ligand structure\nprint('\\nDefining ligand structure...')\nligand_atoms = {\n    'C1': np.array([0.0, 0.0, 0.0]),\n    'C2': np.array([1.5, 0.0, 0.0]),\n    'C3': np.array([2.2, 1.3, 0.0]),\n    'N1': np.array([1.5, 2.6, 0.0]),\n    'O1': np.array([0.0, 2.6, 0.5]),\n    'C4': np.array([-0.7, 1.3, 0.5])\n}\n\nligand_center = np.mean(list(ligand_atoms.values()), axis=0)\nligand_coords = np.array(list(ligand_atoms.values()))\nligand_coords_centered = ligand_coords - ligand_center\n\nprint(f'Ligand atoms: {len(ligand_atoms)}')\nprint(f'Ligand center: ({ligand_center[0]:.1f}, {ligand_center[1]:.1f}, {ligand_center[2]:.1f})')\nprint(f'Ligand size: {np.max(cdist(ligand_coords, ligand_coords)):.1f} Å')\n\n# Simulate diffusion-based pose generation\nprint('\\n=== Diffusion-Based Pose Generation ===')\n\ndef rotation_matrix(axis, angle):\n    \"\"\"Generate rotation matrix for given axis and angle\"\"\"\n    axis = axis / np.linalg.norm(axis)\n    cos_angle = np.cos(angle)\n    sin_angle = np.sin(angle)\n    \n    return (\n        cos_angle * np.eye(3) +\n        sin_angle * np.array([[0, -axis[2], axis[1]],\n                             [axis[2], 0, -axis[0]],\n                             [-axis[1], axis[0], 0]]) +\n        (1 - cos_angle) * np.outer(axis, axis)\n    )\n\ndef generate_pose(translation, rotation_axis, rotation_angle):\n    \"\"\"Generate ligand pose from translation and rotation\"\"\"\n    rot_matrix = rotation_matrix(rotation_axis, rotation_angle)\n    rotated_ligand = np.dot(ligand_coords_centered, rot_matrix.T)\n    posed_ligand = rotated_ligand + translation\n    return posed_ligand\n\ndef calculate_docking_score(posed_ligand, protein_surface):\n    \"\"\"Calculate simplified docking score\"\"\"\n    # Distance-based scoring\n    min_distances = np.min(cdist(posed_ligand, protein_surface), axis=1)\n    \n    # Prefer poses where ligand is close to protein surface\n    distance_score = -np.mean(np.maximum(0, min_distances - 2.0))\n    \n    # Penalty for steric clashes (too close)\n    clash_penalty = -np.sum(np.maximum(0, 1.5 - min_distances)) * 10\n    \n    # Bonus for being in binding site\n    center_distances = np.linalg.norm(posed_ligand - binding_site_center, axis=1)\n    binding_site_bonus = -np.mean(np.maximum(0, center_distances - binding_site_radius))\n    \n    total_score = distance_score + clash_penalty + binding_site_bonus\n    return total_score, distance_score, clash_penalty, binding_site_bonus\n\n# Generate multiple poses using diffusion-like sampling\nprint('Generating poses via diffusion sampling...')\nn_poses = 1000\nposes = []\nscores = []\nscore_components = []\n\nfor i in range(n_poses):\n    # Sample translation (biased toward binding site)\n    translation = binding_site_center + np.random.normal(0, 3.0, 3)\n    \n    # Sample rotation\n    rotation_axis = np.random.normal(0, 1, 3)\n    rotation_axis = rotation_axis / np.linalg.norm(rotation_axis)\n    rotation_angle = np.random.uniform(0, 2*np.pi)\n    \n    # Generate pose\n    posed_ligand = generate_pose(translation, rotation_axis, rotation_angle)\n    \n    # Calculate score\n    total_score, dist_score, clash_penalty, binding_bonus = calculate_docking_score(\n        posed_ligand, protein_surface)\n    \n    poses.append({\n        'id': i,\n        'translation': translation,\n        'rotation_axis': rotation_axis,\n        'rotation_angle': rotation_angle,\n        'coordinates': posed_ligand,\n        'score': total_score,\n        'distance_score': dist_score,\n        'clash_penalty': clash_penalty,\n        'binding_bonus': binding_bonus\n    })\n    \n    scores.append(total_score)\n    score_components.append([dist_score, clash_penalty, binding_bonus])\n\nscores = np.array(scores)\nscore_components = np.array(score_components)\n\nprint(f'Generated {n_poses} poses')\nprint(f'Score range: {np.min(scores):.2f} to {np.max(scores):.2f}')\nprint(f'Mean score: {np.mean(scores):.2f} ± {np.std(scores):.2f}')\n\n# Select top poses\ntop_indices = np.argsort(scores)[-50:]  # Top 50 poses\ntop_poses = [poses[i] for i in top_indices]\ntop_scores = scores[top_indices]\n\nprint(f'\\nTop 10 pose scores: {top_scores[-10:]}')\n\n# Cluster top poses\nprint('\\n=== Pose Clustering ===')\n\n# Extract center positions of top poses\ntop_centers = np.array([np.mean(pose['coordinates'], axis=0) for pose in top_poses])\n\n# Cluster by position\ndbscan = DBSCAN(eps=2.0, min_samples=3)\ncluster_labels = dbscan.fit_predict(top_centers)\n\nn_clusters = len(set(cluster_labels)) - (1 if -1 in cluster_labels else 0)\nn_noise = list(cluster_labels).count(-1)\n\nprint(f'Found {n_clusters} pose clusters')\nprint(f'Noise points: {n_noise}')\n\n# Analyze clusters\ncluster_info = []\nfor cluster_id in range(n_clusters):\n    cluster_mask = cluster_labels == cluster_id\n    cluster_poses = [top_poses[i] for i in range(len(top_poses)) if cluster_mask[i]]\n    cluster_scores = top_scores[cluster_mask]\n    \n    cluster_center = np.mean(top_centers[cluster_mask], axis=0)\n    best_score = np.max(cluster_scores)\n    cluster_size = len(cluster_poses)\n    \n    cluster_info.append({\n        'id': cluster_id,\n        'size': cluster_size,\n        'best_score': best_score,\n        'center': cluster_center,\n        'poses': cluster_poses\n    })\n    \n    print(f'  Cluster {cluster_id}: {cluster_size} poses, '\n          f'best score: {best_score:.2f}, '\n          f'center: ({cluster_center[0]:.1f}, {cluster_center[1]:.1f}, {cluster_center[2]:.1f})')\n\n# Select representative poses\nprint('\\n=== Representative Pose Selection ===')\n\nrepresentative_poses = []\nfor cluster in sorted(cluster_info, key=lambda x: x['best_score'], reverse=True):\n    # Select best pose from each cluster\n    best_pose_idx = np.argmax([pose['score'] for pose in cluster['poses']])\n    best_pose = cluster['poses'][best_pose_idx]\n    \n    representative_poses.append({\n        'cluster_id': cluster['id'],\n        'pose': best_pose,\n        'cluster_size': cluster['size']\n    })\n    \n    print(f'Cluster {cluster[\"id\"]}: Score {best_pose[\"score\"]:.2f}, '\n          f'Position ({np.mean(best_pose[\"coordinates\"], axis=0)})')\n\n# Analyze binding mode\nprint('\\n=== Binding Mode Analysis ===')\n\nif representative_poses:\n    best_overall = representative_poses[0]['pose']\n    \n    print(f'Best pose analysis:')\n    print(f'  Total score: {best_overall[\"score\"]:.2f}')\n    print(f'  Distance score: {best_overall[\"distance_score\"]:.2f}')\n    print(f'  Clash penalty: {best_overall[\"clash_penalty\"]:.2f}')\n    print(f'  Binding site bonus: {best_overall[\"binding_bonus\"]:.2f}')\n    \n    # Calculate interactions\n    best_coords = best_overall['coordinates']\n    distances_to_surface = cdist(best_coords, protein_surface)\n    min_distances = np.min(distances_to_surface, axis=1)\n    \n    print(f'\\n  Ligand-protein distances:')\n    for i, (atom_name, distance) in enumerate(zip(ligand_atoms.keys(), min_distances)):\n        print(f'    {atom_name}: {distance:.2f} Å')\n    \n    # Identify close contacts\n    close_contacts = min_distances < 3.0\n    print(f'\\n  Close contacts (< 3.0 Å): {np.sum(close_contacts)} atoms')\n    \n    # Center of mass analysis\n    com_distance = np.linalg.norm(np.mean(best_coords, axis=0) - binding_site_center)\n    print(f'  Distance from binding site center: {com_distance:.2f} Å')\n\n# Confidence estimation\nprint('\\n=== Confidence Estimation ===')\n\n# Score-based confidence\ntop_10_scores = top_scores[-10:]\nscore_std = np.std(top_10_scores)\nscore_confidence = 1.0 / (1.0 + score_std)  # Higher std = lower confidence\n\n# Clustering-based confidence\nif n_clusters > 0:\n    largest_cluster_size = max(cluster['size'] for cluster in cluster_info)\n    cluster_confidence = largest_cluster_size / len(top_poses)\nelse:\n    cluster_confidence = 0.0\n\n# Combined confidence\noverall_confidence = (score_confidence + cluster_confidence) / 2.0\n\nprint(f'Score-based confidence: {score_confidence:.3f}')\nprint(f'Clustering confidence: {cluster_confidence:.3f}')\nprint(f'Overall confidence: {overall_confidence:.3f}')\n\n# Quality metrics\nprint('\\n=== Quality Metrics ===')\n\n# Pose diversity\nall_centers = np.array([np.mean(pose['coordinates'], axis=0) for pose in poses])\nposition_diversity = np.mean(cdist(all_centers, all_centers))\n\n# Energy landscape analysis\nenergy_range = np.max(scores) - np.min(scores)\nfunnel_quality = (np.max(scores) - np.mean(scores)) / np.std(scores)\n\nprint(f'Position diversity: {position_diversity:.2f} Å')\nprint(f'Energy range: {energy_range:.2f}')\nprint(f'Funnel quality: {funnel_quality:.2f}')\n\n# Visualization\nprint('\\n=== Visualization ===')\n\nfig = plt.figure(figsize=(16, 12))\n\n# 3D visualization of binding site and top poses\nax1 = fig.add_subplot(221, projection='3d')\n\n# Plot protein surface\nax1.scatter(protein_surface[:, 0], protein_surface[:, 1], protein_surface[:, 2], \n           c='lightblue', alpha=0.3, s=20, label='Protein surface')\n\n# Plot binding site center\nax1.scatter(*binding_site_center, c='red', s=100, marker='*', label='Binding site')\n\n# Plot top poses\ncolors = plt.cm.viridis(np.linspace(0, 1, len(representative_poses)))\nfor i, rep_pose in enumerate(representative_poses[:5]):  # Show top 5\n    coords = rep_pose['pose']['coordinates']\n    ax1.scatter(coords[:, 0], coords[:, 1], coords[:, 2], \n               c=[colors[i]], s=50, alpha=0.8, label=f'Pose {i+1}')\n\nax1.set_xlabel('X (Å)')\nax1.set_ylabel('Y (Å)')\nax1.set_zlabel('Z (Å)')\nax1.set_title('3D Binding Site and Top Poses')\nax1.legend()\n\n# Score distribution\nax2 = fig.add_subplot(222)\nax2.hist(scores, bins=50, alpha=0.7, color='skyblue', edgecolor='black')\nax2.axvline(np.mean(scores), color='red', linestyle='--', label=f'Mean: {np.mean(scores):.2f}')\nax2.axvline(np.percentile(scores, 95), color='green', linestyle='--', \n           label=f'95th percentile: {np.percentile(scores, 95):.2f}')\nax2.set_xlabel('Docking Score')\nax2.set_ylabel('Frequency')\nax2.set_title('Score Distribution')\nax2.legend()\n\n# Score components\nax3 = fig.add_subplot(223)\ncomponent_names = ['Distance', 'Clash Penalty', 'Binding Bonus']\nfor i, name in enumerate(component_names):\n    ax3.scatter(range(len(top_poses)), score_components[top_indices, i], \n               alpha=0.6, label=name, s=20)\nax3.set_xlabel('Pose Index (sorted by score)')\nax3.set_ylabel('Score Component')\nax3.set_title('Score Components for Top Poses')\nax3.legend()\nax3.grid(True, alpha=0.3)\n\n# Cluster analysis\nax4 = fig.add_subplot(224)\nif n_clusters > 0:\n    for cluster_id in range(n_clusters):\n        cluster_mask = cluster_labels == cluster_id\n        if np.any(cluster_mask):\n            cluster_centers = top_centers[cluster_mask]\n            cluster_scores = top_scores[cluster_mask]\n            ax4.scatter(cluster_centers[:, 0], cluster_centers[:, 1], \n                       c=cluster_scores, s=50, alpha=0.7, \n                       label=f'Cluster {cluster_id}', cmap='viridis')\n    \n    # Noise points\n    noise_mask = cluster_labels == -1\n    if np.any(noise_mask):\n        ax4.scatter(top_centers[noise_mask, 0], top_centers[noise_mask, 1], \n                   c='gray', s=20, alpha=0.5, label='Noise')\n    \n    ax4.scatter(*binding_site_center[:2], c='red', s=100, marker='*', \n               label='Binding site')\n    \n    ax4.set_xlabel('X (Å)')\n    ax4.set_ylabel('Y (Å)')\n    ax4.set_title('Pose Clusters (XY projection)')\n    ax4.legend()\n    ax4.grid(True, alpha=0.3)\n\nplt.tight_layout()\n\n# Save visualization\nwith tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:\n    plt.savefig(tmp.name, dpi=150, bbox_inches='tight')\n    viz_file = tmp.name\n\nplt.close()\nprint(f'Docking visualization saved to: {viz_file}')\n\n# Summary report\nprint('\\n' + '=' * 50)\nprint('DIFFDOCK MOLECULAR DOCKING SUMMARY')\nprint('=' * 50)\nprint(f'Poses generated: {n_poses:,}')\nprint(f'Top poses analyzed: {len(top_poses)}')\nprint(f'Pose clusters found: {n_clusters}')\nprint(f'Best docking score: {np.max(scores):.2f}')\nprint(f'Confidence score: {overall_confidence:.3f}')\nprint(f'\\nBest pose location:')\nif representative_poses:\n    best_com = np.mean(representative_poses[0]['pose']['coordinates'], axis=0)\n    print(f'  Center of mass: ({best_com[0]:.2f}, {best_com[1]:.2f}, {best_com[2]:.2f})')\n    print(f'  Distance from binding site: {np.linalg.norm(best_com - binding_site_center):.2f} Å')\n\nprint(f'\\nQuality metrics:')\nprint(f'  Position diversity: {position_diversity:.2f} Å')\nprint(f'  Energy range: {energy_range:.2f}')\nprint(f'  Funnel quality: {funnel_quality:.2f}')\n\n# Cleanup\nos.unlink(viz_file)\nprint('\\nDemo complete - temporary files cleaned up')\n\nprint('\\nDiffDock provides:')\nprint('• State-of-the-art docking accuracy')\nprint('• End-to-end deep learning approach')\nprint('• Fast GPU-accelerated inference')\nprint('• Handling of flexible ligands and proteins')\nprint('• Confidence estimation for poses')\nprint('• Integration with drug discovery pipelines')\nprint('• Pre-trained models on large datasets')\n\nprint('\\nTypical DiffDock usage:')\nprint('python -m diffdock.inference --protein protein.pdb \\\\')\nprint('  --ligand ligand.sdf --out_dir results --inference_steps 20')",
+      "usage_example": "# DiffDock molecular docking simulation\n# This demonstrates the concepts and workflow\n\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom scipy.spatial.distance import cdist\nfrom sklearn.cluster import DBSCAN\nimport tempfile\nimport os\n\nprint('DiffDock - Diffusion Model for Molecular Docking')\nprint('=' * 50)\n\n# Overview of DiffDock approach\nprint('DiffDock Innovation:')\nprint('• Uses diffusion models for pose generation')\nprint('• End-to-end learning without hand-crafted features')\nprint('• Handles both rigid and flexible docking')\nprint('• State-of-the-art accuracy on benchmarks')\nprint('• Fast inference with GPU acceleration')\n\nprint('\\nKey Advantages over Traditional Docking:')\nprint('• No need for extensive sampling')\nprint('• Better handling of conformational flexibility')\nprint('• Learned representations from large datasets')\nprint('• Robust to protein structure variations')\n\n# Simulate molecular docking workflow\nprint('\\n=== Molecular Docking Simulation ===')\n\nnp.random.seed(42)\n\n# Define protein binding site\nprint('Defining protein binding site...')\nbinding_site_center = np.array([0.0, 0.0, 0.0])\nbinding_site_radius = 8.0\n\n# Generate protein surface points\nn_surface_points = 200\ntheta = np.random.uniform(0, 2*np.pi, n_surface_points)\nphi = np.random.uniform(0, np.pi, n_surface_points)\nr = np.random.uniform(5.0, 8.0, n_surface_points)\n\nprotein_surface = np.column_stack([\n    r * np.sin(phi) * np.cos(theta),\n    r * np.sin(phi) * np.sin(theta),\n    r * np.cos(phi)\n])\n\nprint(f'Generated {n_surface_points} protein surface points')\nprint(f'Binding site radius: {binding_site_radius:.1f} Å')\n\n# Define ligand structure\nprint('\\nDefining ligand structure...')\nligand_atoms = {\n    'C1': np.array([0.0, 0.0, 0.0]),\n    'C2': np.array([1.5, 0.0, 0.0]),\n    'C3': np.array([2.2, 1.3, 0.0]),\n    'N1': np.array([1.5, 2.6, 0.0]),\n    'O1': np.array([0.0, 2.6, 0.5]),\n    'C4': np.array([-0.7, 1.3, 0.5])\n}\n\nligand_center = np.mean(list(ligand_atoms.values()), axis=0)\nligand_coords = np.array(list(ligand_atoms.values()))\nligand_coords_centered = ligand_coords - ligand_center\n\nprint(f'Ligand atoms: {len(ligand_atoms)}')\nprint(f'Ligand center: ({ligand_center[0]:.1f}, {ligand_center[1]:.1f}, {ligand_center[2]:.1f})')\nprint(f'Ligand size: {np.max(cdist(ligand_coords, ligand_coords)):.1f} Å')\n\n# Simulate diffusion-based pose generation\nprint('\\n=== Diffusion-Based Pose Generation ===')\n\ndef rotation_matrix(axis, angle):\n    \"\"\"Generate rotation matrix for given axis and angle\"\"\"\n    axis = axis / np.linalg.norm(axis)\n    cos_angle = np.cos(angle)\n    sin_angle = np.sin(angle)\n    \n    return (\n        cos_angle * np.eye(3) +\n        sin_angle * np.array([[0, -axis[2], axis[1]],\n                             [axis[2], 0, -axis[0]],\n                             [-axis[1], axis[0], 0]]) +\n        (1 - cos_angle) * np.outer(axis, axis)\n    )\n\ndef generate_pose(translation, rotation_axis, rotation_angle):\n    \"\"\"Generate ligand pose from translation and rotation\"\"\"\n    rot_matrix = rotation_matrix(rotation_axis, rotation_angle)\n    rotated_ligand = np.dot(ligand_coords_centered, rot_matrix.T)\n    posed_ligand = rotated_ligand + translation\n    return posed_ligand\n\ndef calculate_docking_score(posed_ligand, protein_surface):\n    \"\"\"Calculate simplified docking score\"\"\"\n    # Distance-based scoring\n    min_distances = np.min(cdist(posed_ligand, protein_surface), axis=1)\n    \n    # Prefer poses where ligand is close to protein surface\n    distance_score = -np.mean(np.maximum(0, min_distances - 2.0))\n    \n    # Penalty for steric clashes (too close)\n    clash_penalty = -np.sum(np.maximum(0, 1.5 - min_distances)) * 10\n    \n    # Bonus for being in binding site\n    center_distances = np.linalg.norm(posed_ligand - binding_site_center, axis=1)\n    binding_site_bonus = -np.mean(np.maximum(0, center_distances - binding_site_radius))\n    \n    total_score = distance_score + clash_penalty + binding_site_bonus\n    return total_score, distance_score, clash_penalty, binding_site_bonus\n\n# Generate multiple poses using diffusion-like sampling\nprint('Generating poses via diffusion sampling...')\nn_poses = 1000\nposes = []\nscores = []\nscore_components = []\n\nfor i in range(n_poses):\n    # Sample translation (biased toward binding site)\n    translation = binding_site_center + np.random.normal(0, 3.0, 3)\n    \n    # Sample rotation\n    rotation_axis = np.random.normal(0, 1, 3)\n    rotation_axis = rotation_axis / np.linalg.norm(rotation_axis)\n    rotation_angle = np.random.uniform(0, 2*np.pi)\n    \n    # Generate pose\n    posed_ligand = generate_pose(translation, rotation_axis, rotation_angle)\n    \n    # Calculate score\n    total_score, dist_score, clash_penalty, binding_bonus = calculate_docking_score(\n        posed_ligand, protein_surface)\n    \n    poses.append({\n        'id': i,\n        'translation': translation,\n        'rotation_axis': rotation_axis,\n        'rotation_angle': rotation_angle,\n        'coordinates': posed_ligand,\n        'score': total_score,\n        'distance_score': dist_score,\n        'clash_penalty': clash_penalty,\n        'binding_bonus': binding_bonus\n    })\n    \n    scores.append(total_score)\n    score_components.append([dist_score, clash_penalty, binding_bonus])\n\nscores = np.array(scores)\nscore_components = np.array(score_components)\n\nprint(f'Generated {n_poses} poses')\nprint(f'Score range: {np.min(scores):.2f} to {np.max(scores):.2f}')\nprint(f'Mean score: {np.mean(scores):.2f} ± {np.std(scores):.2f}')\n\n# Select top poses\ntop_indices = np.argsort(scores)[-50:]  # Top 50 poses\ntop_poses = [poses[i] for i in top_indices]\ntop_scores = scores[top_indices]\n\nprint(f'\\nTop 10 pose scores: {top_scores[-10:]}')\n\n# Cluster top poses\nprint('\\n=== Pose Clustering ===')\n\n# Extract center positions of top poses\ntop_centers = np.array([np.mean(pose['coordinates'], axis=0) for pose in top_poses])\n\n# Cluster by position\ndbscan = DBSCAN(eps=2.0, min_samples=3)\ncluster_labels = dbscan.fit_predict(top_centers)\n\nn_clusters = len(set(cluster_labels)) - (1 if -1 in cluster_labels else 0)\nn_noise = list(cluster_labels).count(-1)\n\nprint(f'Found {n_clusters} pose clusters')\nprint(f'Noise points: {n_noise}')\n\n# Analyze clusters\ncluster_info = []\nfor cluster_id in range(n_clusters):\n    cluster_mask = cluster_labels == cluster_id\n    cluster_poses = [top_poses[i] for i in range(len(top_poses)) if cluster_mask[i]]\n    cluster_scores = top_scores[cluster_mask]\n    \n    cluster_center = np.mean(top_centers[cluster_mask], axis=0)\n    best_score = np.max(cluster_scores)\n    cluster_size = len(cluster_poses)\n    \n    cluster_info.append({\n        'id': cluster_id,\n        'size': cluster_size,\n        'best_score': best_score,\n        'center': cluster_center,\n        'poses': cluster_poses\n    })\n    \n    print(f'  Cluster {cluster_id}: {cluster_size} poses, '\n          f'best score: {best_score:.2f}, '\n          f'center: ({cluster_center[0]:.1f}, {cluster_center[1]:.1f}, {cluster_center[2]:.1f})')\n\n# Select representative poses\nprint('\\n=== Representative Pose Selection ===')\n\nrepresentative_poses = []\nfor cluster in sorted(cluster_info, key=lambda x: x['best_score'], reverse=True):\n    # Select best pose from each cluster\n    best_pose_idx = np.argmax([pose['score'] for pose in cluster['poses']])\n    best_pose = cluster['poses'][best_pose_idx]\n    \n    representative_poses.append({\n        'cluster_id': cluster['id'],\n        'pose': best_pose,\n        'cluster_size': cluster['size']\n    })\n    \n    print(f'Cluster {cluster[\"id\"]}: Score {best_pose[\"score\"]:.2f}, '\n          f'Position ({np.mean(best_pose[\"coordinates\"], axis=0)})')\n\n# Analyze binding mode\nprint('\\n=== Binding Mode Analysis ===')\n\nif representative_poses:\n    best_overall = representative_poses[0]['pose']\n    \n    print(f'Best pose analysis:')\n    print(f'  Total score: {best_overall[\"score\"]:.2f}')\n    print(f'  Distance score: {best_overall[\"distance_score\"]:.2f}')\n    print(f'  Clash penalty: {best_overall[\"clash_penalty\"]:.2f}')\n    print(f'  Binding site bonus: {best_overall[\"binding_bonus\"]:.2f}')\n    \n    # Calculate interactions\n    best_coords = best_overall['coordinates']\n    distances_to_surface = cdist(best_coords, protein_surface)\n    min_distances = np.min(distances_to_surface, axis=1)\n    \n    print(f'\\n  Ligand-protein distances:')\n    for i, (atom_name, distance) in enumerate(zip(ligand_atoms.keys(), min_distances)):\n        print(f'    {atom_name}: {distance:.2f} Å')\n    \n    # Identify close contacts\n    close_contacts = min_distances < 3.0\n    print(f'\\n  Close contacts (< 3.0 Å): {np.sum(close_contacts)} atoms')\n    \n    # Center of mass analysis\n    com_distance = np.linalg.norm(np.mean(best_coords, axis=0) - binding_site_center)\n    print(f'  Distance from binding site center: {com_distance:.2f} Å')\n\n# Confidence estimation\nprint('\\n=== Confidence Estimation ===')\n\n# Score-based confidence\ntop_10_scores = top_scores[-10:]\nscore_std = np.std(top_10_scores)\nscore_confidence = 1.0 / (1.0 + score_std)  # Higher std = lower confidence\n\n# Clustering-based confidence\nif n_clusters > 0:\n    largest_cluster_size = max(cluster['size'] for cluster in cluster_info)\n    cluster_confidence = largest_cluster_size / len(top_poses)\nelse:\n    cluster_confidence = 0.0\n\n# Combined confidence\noverall_confidence = (score_confidence + cluster_confidence) / 2.0\n\nprint(f'Score-based confidence: {score_confidence:.3f}')\nprint(f'Clustering confidence: {cluster_confidence:.3f}')\nprint(f'Overall confidence: {overall_confidence:.3f}')\n\n# Quality metrics\nprint('\\n=== Quality Metrics ===')\n\n# Pose diversity\nall_centers = np.array([np.mean(pose['coordinates'], axis=0) for pose in poses])\nposition_diversity = np.mean(cdist(all_centers, all_centers))\n\n# Energy landscape analysis\nenergy_range = np.max(scores) - np.min(scores)\nfunnel_quality = (np.max(scores) - np.mean(scores)) / np.std(scores)\n\nprint(f'Position diversity: {position_diversity:.2f} Å')\nprint(f'Energy range: {energy_range:.2f}')\nprint(f'Funnel quality: {funnel_quality:.2f}')\n\n# Visualization\nprint('\\n=== Visualization ===')\n\nfig = plt.figure(figsize=(16, 12))\n\n# 3D visualization of binding site and top poses\nax1 = fig.add_subplot(221, projection='3d')\n\n# Plot protein surface\nax1.scatter(protein_surface[:, 0], protein_surface[:, 1], protein_surface[:, 2], \n           c='lightblue', alpha=0.3, s=20, label='Protein surface')\n\n# Plot binding site center\nax1.scatter(*binding_site_center, c='red', s=100, marker='*', label='Binding site')\n\n# Plot top poses\ncolors = plt.cm.viridis(np.linspace(0, 1, len(representative_poses)))\nfor i, rep_pose in enumerate(representative_poses[:5]):  # Show top 5\n    coords = rep_pose['pose']['coordinates']\n    ax1.scatter(coords[:, 0], coords[:, 1], coords[:, 2], \n               c=[colors[i]], s=50, alpha=0.8, label=f'Pose {i+1}')\n\nax1.set_xlabel('X (Å)')\nax1.set_ylabel('Y (Å)')\nax1.set_zlabel('Z (Å)')\nax1.set_title('3D Binding Site and Top Poses')\nax1.legend()\n\n# Score distribution\nax2 = fig.add_subplot(222)\nax2.hist(scores, bins=50, alpha=0.7, color='skyblue', edgecolor='black')\nax2.axvline(np.mean(scores), color='red', linestyle='--', label=f'Mean: {np.mean(scores):.2f}')\nax2.axvline(np.percentile(scores, 95), color='green', linestyle='--', \n           label=f'95th percentile: {np.percentile(scores, 95):.2f}')\nax2.set_xlabel('Docking Score')\nax2.set_ylabel('Frequency')\nax2.set_title('Score Distribution')\nax2.legend()\n\n# Score components\nax3 = fig.add_subplot(223)\ncomponent_names = ['Distance', 'Clash Penalty', 'Binding Bonus']\nfor i, name in enumerate(component_names):\n    ax3.scatter(range(len(top_poses)), score_components[top_indices, i], \n               alpha=0.6, label=name, s=20)\nax3.set_xlabel('Pose Index (sorted by score)')\nax3.set_ylabel('Score Component')\nax3.set_title('Score Components for Top Poses')\nax3.legend()\nax3.grid(True, alpha=0.3)\n\n# Cluster analysis\nax4 = fig.add_subplot(224)\nif n_clusters > 0:\n    for cluster_id in range(n_clusters):\n        cluster_mask = cluster_labels == cluster_id\n        if np.any(cluster_mask):\n            cluster_centers = top_centers[cluster_mask]\n            cluster_scores = top_scores[cluster_mask]\n            ax4.scatter(cluster_centers[:, 0], cluster_centers[:, 1], \n                       c=cluster_scores, s=50, alpha=0.7, \n                       label=f'Cluster {cluster_id}', cmap='viridis')\n    \n    # Noise points\n    noise_mask = cluster_labels == -1\n    if np.any(noise_mask):\n        ax4.scatter(top_centers[noise_mask, 0], top_centers[noise_mask, 1], \n                   c='gray', s=20, alpha=0.5, label='Noise')\n    \n    ax4.scatter(*binding_site_center[:2], c='red', s=100, marker='*', \n               label='Binding site')\n    \n    ax4.set_xlabel('X (Å)')\n    ax4.set_ylabel('Y (Å)')\n    ax4.set_title('Pose Clusters (XY projection)')\n    ax4.legend()\n    ax4.grid(True, alpha=0.3)\n\nplt.tight_layout()\n\n# Save visualization\nwith tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:\n    plt.savefig(tmp.name, dpi=150, bbox_inches='tight')\n    viz_file = tmp.name\n\nplt.close()\nprint(f'Docking visualization saved to: {viz_file}')\n\n# Summary report\nprint('\\n' + '=' * 50)\nprint('DIFFDOCK MOLECULAR DOCKING SUMMARY')\nprint('=' * 50)\nprint(f'Poses generated: {n_poses:}')\nprint(f'Top poses analyzed: {len(top_poses)}')\nprint(f'Pose clusters found: {n_clusters}')\nprint(f'Best docking score: {np.max(scores):.2f}')\nprint(f'Confidence score: {overall_confidence:.3f}')\nprint(f'\\nBest pose location:')\nif representative_poses:\n    best_com = np.mean(representative_poses[0]['pose']['coordinates'], axis=0)\n    print(f'  Center of mass: ({best_com[0]:.2f}, {best_com[1]:.2f}, {best_com[2]:.2f})')\n    print(f'  Distance from binding site: {np.linalg.norm(best_com - binding_site_center):.2f} Å')\n\nprint(f'\\nQuality metrics:')\nprint(f'  Position diversity: {position_diversity:.2f} Å')\nprint(f'  Energy range: {energy_range:.2f}')\nprint(f'  Funnel quality: {funnel_quality:.2f}')\n\n# Cleanup\nos.unlink(viz_file)\nprint('\\nDemo complete - temporary files cleaned up')\n\nprint('\\nDiffDock provides:')\nprint('• State-of-the-art docking accuracy')\nprint('• End-to-end deep learning approach')\nprint('• Fast GPU-accelerated inference')\nprint('• Handling of flexible ligands and proteins')\nprint('• Confidence estimation for poses')\nprint('• Integration with drug discovery pipelines')\nprint('• Pre-trained models on large datasets')\n\nprint('\\nTypical DiffDock usage:')\nprint('python -m diffdock.inference --protein protein.pdb \\\\')\nprint('  --ligand ligand.sdf --out_dir results --inference_steps 20')",
       "quick_start": [
         "Install: pip install diffdock",
         "Prepare: protein.pdb and ligand.sdf files",

tooluniverse/data/pmc_tools.json

@@ -32,10 +32,7 @@
       },
       "required": [
         "query",
-        "limit",
-        "date_from",
-        "date_to",
-        "article_type"
+        "limit"
       ]
     },
     "return_schema": {

tooluniverse/data/ppi_tools.json

@@ -0,0 +1,139 @@
+[
+  {
+    "type": "STRINGRESTTool",
+    "name": "STRING_get_protein_interactions",
+    "description": "Query protein-protein interactions from the STRING database. STRING is a comprehensive database of known and predicted protein-protein interactions with confidence scores and functional annotations.",
+    "parameter": {
+      "type": "object",
+      "properties": {
+        "protein_ids": {
+          "type": "array",
+          "items": {"type": "string"},
+          "description": "List of protein identifiers (UniProt IDs, gene names, etc.)",
+          "minItems": 1
+        },
+        "species": {
+          "type": "integer",
+          "description": "NCBI taxonomy ID (default: 9606 for human)",
+          "default": 9606
+        },
+        "confidence_score": {
+          "type": "number",
+          "description": "Minimum confidence score (0-1, default: 0.4)",
+          "minimum": 0,
+          "maximum": 1,
+          "default": 0.4
+        },
+        "limit": {
+          "type": "integer",
+          "description": "Maximum number of interactions to return (default: 50)",
+          "minimum": 1,
+          "maximum": 1000,
+          "default": 50
+        },
+        "network_type": {
+          "type": "string",
+          "description": "Type of network ('full', 'physical', 'functional')",
+          "enum": ["full", "physical", "functional"],
+          "default": "full"
+        }
+      },
+      "required": ["protein_ids"]
+    },
+    "fields": {
+      "endpoint": "/tsv/network",
+      "return_format": "TSV"
+    },
+    "return_schema": {
+      "type": "object",
+      "properties": {
+        "success": {"type": "boolean"},
+        "interactions": {
+          "type": "array",
+          "items": {
+            "type": "object",
+            "properties": {
+              "protein1": {"type": "string"},
+              "protein2": {"type": "string"},
+              "protein1_name": {"type": "string"},
+              "protein2_name": {"type": "string"},
+              "combined_score": {"type": "number"},
+              "confidence_score": {"type": "number"},
+              "interaction_sources": {"type": "array", "items": {"type": "string"}},
+              "interaction_type": {"type": "string"}
+            }
+          }
+        },
+        "mapping_data": {"type": "array"},
+        "functional_enrichment": {"type": "array"},
+        "summary": {"type": "object"},
+        "error": {"type": "string"}
+      }
+    },
+    "implementation": {
+      "language": "python",
+      "dependencies": ["requests"],
+      "source_file": "STRING_get_protein_interactions.py"
+    }
+  },
+  {
+    "type": "BioGRIDRESTTool",
+    "name": "BioGRID_get_interactions",
+    "description": "Query protein and genetic interactions from the BioGRID database. BioGRID is a comprehensive database of physical and genetic interactions with detailed experimental evidence.",
+    "parameter": {
+      "type": "object",
+      "properties": {
+        "gene_names": {
+          "type": "array",
+          "items": {"type": "string"},
+          "description": "List of gene names or protein identifiers",
+          "minItems": 1
+        },
+        "organism": {
+          "type": "string",
+          "description": "Organism name (e.g., 'Homo sapiens', 'Mus musculus')",
+          "default": "Homo sapiens"
+        },
+        "interaction_type": {
+          "type": "string",
+          "description": "Type of interaction ('physical', 'genetic', 'both')",
+          "enum": ["physical", "genetic", "both"],
+          "default": "both"
+        },
+        "evidence_types": {
+          "type": "array",
+          "items": {"type": "string"},
+          "description": "List of evidence types to include"
+        },
+        "limit": {
+          "type": "integer",
+          "description": "Maximum number of interactions to return (default: 100)",
+          "minimum": 1,
+          "maximum": 1000,
+          "default": 100
+        },
+        "format": {
+          "type": "string",
+          "description": "Output format ('json' or 'tab', default: 'json')",
+          "enum": ["json", "tab"],
+          "default": "json"
+        }
+      },
+      "required": ["gene_names"]
+    },
+    "return_schema": {
+      "type": "object",
+      "properties": {
+        "success": {"type": "boolean"},
+        "interactions": {"type": "array"},
+        "summary": {"type": "object"},
+        "statistics": {"type": "object"},
+        "error": {"type": "string"}
+      }
+    },
+    "fields": {
+      "endpoint": "/interactions/",
+      "return_format": "JSON"
+    }
+  }
+]

tooluniverse/data/pubmed_tools.json

@@ -21,9 +21,7 @@
         }
       },
       "required": [
-        "query",
-        "limit",
-        "api_key"
+        "query"
       ]
     },
     "return_schema": {

tooluniverse/data/semantic_scholar_tools.json

@@ -22,8 +22,7 @@
       },
       "required": [
         "query",
-        "limit",
-        "api_key"
+        "limit"
       ]
     },
     "return_schema": {

tooluniverse/data/tool_composition_tools.json

@@ -14,13 +14,11 @@
       "properties": {
         "source_tool": {
           "type": "string",
-          "description": "The source tool specification (JSON string with name, description, parameter schema, and example outputs)",
-          "required": true
+          "description": "The source tool specification (JSON string with name, description, parameter schema, and example outputs)"
         },
         "target_tool": {
           "type": "string",
-          "description": "The target tool specification (JSON string with name, description, parameter schema)",
-          "required": true
+          "description": "The target tool specification (JSON string with name, description, parameter schema)"
         },
         "analysis_depth": {
           "type": "string",