workbench 0.8.171__py3-none-any.whl → 0.8.173__py3-none-any.whl

workbench/model_scripts/custom_models/chem_info/mol_descriptors.py

@@ -0,0 +1,471 @@
+"""
+mol_descriptors.py - Molecular descriptor computation for ADMET modeling
+
+Purpose:
+    Computes comprehensive molecular descriptors for ADMET (Absorption, Distribution,
+    Metabolism, Excretion, Toxicity) property prediction. Combines RDKit's full
+    descriptor set with selected Mordred descriptors and custom stereochemistry features.
+
+Descriptor Categories:
+    1. RDKit Descriptors (~220 descriptors)
+       - Constitutional (MW, heavy atom count, rotatable bonds)
+       - Topological (Balaban J, Kappa indices, Chi indices)
+       - Geometric (radius of gyration, spherocity)
+       - Electronic (HOMO/LUMO estimates, partial charges)
+       - Lipophilicity (LogP, MolLogP)
+       - Pharmacophore (H-bond donors/acceptors, aromatic rings)
+       - ADMET-specific (TPSA, QED, Lipinski descriptors)
+
+    2. Mordred Descriptors (~80 descriptors from 5 ADMET-relevant modules)
+       - AcidBase module: pH-dependent properties (nAcid, nBase)
+       - Aromatic module: CYP metabolism features (nAromAtom, nAromBond)
+       - Constitutional module: Structural complexity (~40 descriptors including nSpiro, nBridgehead)
+       - Chi module: Molecular connectivity indices (~42 descriptors, Chi0-Chi4 variants)
+       - CarbonTypes module: Carbon hybridization states for metabolism (~20 descriptors)
+
+    3. Stereochemistry Features (10 custom descriptors)
+       - Stereocenter counts (R/S, defined/undefined)
+       - Stereobond counts (E/Z, defined/undefined)
+       - Stereochemical complexity and coverage metrics
+       - Critical for distinguishing drug enantiomers/diastereomers
+
+Pipeline Integration:
+    This module expects standardized SMILES from mol_standardize.py:
+
+    1. Standardize structures (mol_standardize.py)
+       ↓
+    2. Compute descriptors (this module)
+       ↓
+    3. Feature selection/ML modeling
+
+Output:
+    Returns input DataFrame with added descriptor columns:
+    - ~220 RDKit descriptors
+    - ~85 Mordred descriptors (from 5 modules)
+    - 10 stereochemistry descriptors
+    Total: ~310 descriptors
+
+    Invalid molecules receive NaN values for all descriptors.
+
+Performance Notes:
+    - RDKit descriptors: Fast, vectorized computation
+    - Mordred descriptors: Moderate speed
+    - Stereochemistry: Moderate speed, requires CIP labeling
+    - Memory: <1GB per 10,000 molecules with all descriptors
+
+Special Considerations:
+    - Ipc descriptor excluded due to potential overflow issues
+    - Molecules failing descriptor calculation get NaN (not dropped)
+    - Stereochemistry features optional for non-chiral datasets
+    - Salt information from standardization not included in descriptors
+      (use separately as categorical feature if needed)
+    - Feature selection recommended due to descriptor redundancy
+
+Example Usage:
+    import pandas as pd
+    from mol_standardize import standardize_dataframe
+    from mol_descriptors import compute_descriptors
+
+    # Standard pipeline
+    df = pd.read_csv("molecules.csv")
+    df = standardize_dataframe(df)  # Standardize first
+    df = compute_descriptors(df)    # Then compute descriptors
+
+    # For achiral molecules (faster)
+    df = compute_descriptors(df, include_stereo=False)
+
+    # Custom SMILES column
+    df = compute_descriptors(df, smiles_column='canonical_smiles')
+
+    # The resulting DataFrame is ready for ML modeling
+    X = df.select_dtypes(include=[np.number])  # All numeric descriptors
+    y = df['activity']  # Your target variable
+
+References:
+    - RDKit descriptors: https://www.rdkit.org/docs/GettingStartedInPython.html#descriptors
+    - Mordred: https://github.com/mordred-descriptor/mordred
+    - Stereochemistry in drug discovery: https://doi.org/10.1021/acs.jmedchem.0c00915
+"""
+
+import logging
+import pandas as pd
+import numpy as np
+import re
+from rdkit import Chem
+from rdkit.Chem import Descriptors, rdCIPLabeler
+from rdkit.ML.Descriptors import MoleculeDescriptors
+from mordred import Calculator as MordredCalculator
+from mordred import AcidBase, Aromatic, Constitutional, Chi, CarbonTypes
+
+logger = logging.getLogger("workbench")
+logger.setLevel(logging.DEBUG)
+
+
+def compute_stereochemistry_features(mol):
+    """
+    Compute stereochemistry descriptors using modern RDKit methods.
+
+    Returns dict with 10 stereochemistry descriptors commonly used in ADMET.
+    """
+    if mol is None:
+        return {
+            "num_stereocenters": np.nan,
+            "num_unspecified_stereocenters": np.nan,
+            "num_defined_stereocenters": np.nan,
+            "num_r_centers": np.nan,
+            "num_s_centers": np.nan,
+            "num_stereobonds": np.nan,
+            "num_e_bonds": np.nan,
+            "num_z_bonds": np.nan,
+            "stereo_complexity": np.nan,
+            "frac_defined_stereo": np.nan,
+        }
+
+    try:
+        # Find all potential stereogenic elements
+        stereo_info = Chem.FindPotentialStereo(mol)
+
+        # Initialize counters
+        defined_centers = 0
+        undefined_centers = 0
+        r_centers = 0
+        s_centers = 0
+        defined_bonds = 0
+        undefined_bonds = 0
+        e_bonds = 0
+        z_bonds = 0
+
+        # Assign CIP labels for accurate R/S and E/Z determination
+        rdCIPLabeler.AssignCIPLabels(mol)
+
+        # Process stereogenic elements
+        for element in stereo_info:
+            if element.type == Chem.StereoType.Atom_Tetrahedral:
+                if element.specified == Chem.StereoSpecified.Specified:
+                    defined_centers += 1
+                    # Get the atom and check its CIP code
+                    atom = mol.GetAtomWithIdx(element.centeredOn)
+                    if atom.HasProp("_CIPCode"):
+                        cip = atom.GetProp("_CIPCode")
+                        if cip == "R":
+                            r_centers += 1
+                        elif cip == "S":
+                            s_centers += 1
+                else:
+                    undefined_centers += 1
+
+            elif element.type == Chem.StereoType.Bond_Double:
+                if element.specified == Chem.StereoSpecified.Specified:
+                    defined_bonds += 1
+                    # Get the bond and check its CIP code
+                    bond = mol.GetBondWithIdx(element.centeredOn)
+                    if bond.HasProp("_CIPCode"):
+                        cip = bond.GetProp("_CIPCode")
+                        if cip == "E":
+                            e_bonds += 1
+                        elif cip == "Z":
+                            z_bonds += 1
+                else:
+                    undefined_bonds += 1
+
+        # Calculate derived metrics
+        total_stereocenters = defined_centers + undefined_centers
+        total_stereobonds = defined_bonds + undefined_bonds
+        total_stereo = total_stereocenters + total_stereobonds
+
+        # Stereochemical complexity (total stereogenic elements)
+        stereo_complexity = total_stereo
+
+        # Fraction of defined stereochemistry
+        if total_stereo > 0:
+            frac_defined = (defined_centers + defined_bonds) / total_stereo
+        else:
+            frac_defined = 1.0  # No stereo elements = fully defined
+
+        return {
+            "num_stereocenters": total_stereocenters,
+            "num_unspecified_stereocenters": undefined_centers,
+            "num_defined_stereocenters": defined_centers,
+            "num_r_centers": r_centers,
+            "num_s_centers": s_centers,
+            "num_stereobonds": total_stereobonds,
+            "num_e_bonds": e_bonds,
+            "num_z_bonds": z_bonds,
+            "stereo_complexity": stereo_complexity,
+            "frac_defined_stereo": frac_defined,
+        }
+
+    except Exception as e:
+        logger.warning(f"Stereochemistry calculation failed: {e}")
+        return {
+            "num_stereocenters": np.nan,
+            "num_unspecified_stereocenters": np.nan,
+            "num_defined_stereocenters": np.nan,
+            "num_r_centers": np.nan,
+            "num_s_centers": np.nan,
+            "num_stereobonds": np.nan,
+            "num_e_bonds": np.nan,
+            "num_z_bonds": np.nan,
+            "stereo_complexity": np.nan,
+            "frac_defined_stereo": np.nan,
+        }
+
+
+def compute_descriptors(df: pd.DataFrame, include_mordred: bool = True, include_stereo: bool = True) -> pd.DataFrame:
+    """
+    Compute all molecular descriptors for ADMET modeling.
+
+    Args:
+        df: Input DataFrame with SMILES
+        include_mordred: Whether to compute Mordred descriptors (default True)
+        include_stereo: Whether to compute stereochemistry features (default True)
+
+    Returns:
+        DataFrame with all descriptor columns added
+
+    Example:
+        df = standardize(df)  # First standardize
+        df = compute_descriptors(df)    # Then compute descriptors with stereo
+        df = compute_descriptors(df, include_stereo=False)  # Without stereo
+        df = compute_descriptors(df, include_mordred=False)  # RDKit only
+    """
+
+    # Check for the smiles column (any capitalization)
+    smiles_column = next((col for col in df.columns if col.lower() == "smiles"), None)
+    if smiles_column is None:
+        raise ValueError("Input DataFrame must have a 'smiles' column")
+
+    result = df.copy()
+
+    # Create molecule objects
+    logger.info("Creating molecule objects...")
+    molecules = []
+    for idx, row in result.iterrows():
+        smiles = row[smiles_column]
+
+        if pd.isna(smiles) or smiles == "":
+            molecules.append(None)
+        else:
+            mol = Chem.MolFromSmiles(smiles)
+            molecules.append(mol)
+
+    # Compute RDKit descriptors
+    logger.info("Computing RDKit Descriptors...")
+
+    # Get all RDKit descriptors
+    all_descriptors = [x[0] for x in Descriptors._descList]
+
+    # Remove IPC descriptor due to overflow issue
+    # See: https://github.com/rdkit/rdkit/issues/1527
+    if "Ipc" in all_descriptors:
+        all_descriptors.remove("Ipc")
+
+    # Make sure we don't have duplicates
+    all_descriptors = list(set(all_descriptors))
+
+    # Initialize calculator
+    calc = MoleculeDescriptors.MolecularDescriptorCalculator(all_descriptors)
+
+    # Compute descriptors
+    descriptor_values = []
+    for mol in molecules:
+        if mol is None:
+            descriptor_values.append([np.nan] * len(all_descriptors))
+        else:
+            try:
+                values = calc.CalcDescriptors(mol)
+                descriptor_values.append(values)
+            except Exception as e:
+                logger.warning(f"RDKit descriptor calculation failed: {e}")
+                descriptor_values.append([np.nan] * len(all_descriptors))
+
+    # Create RDKit features DataFrame
+    rdkit_features_df = pd.DataFrame(descriptor_values, columns=calc.GetDescriptorNames(), index=result.index)
+
+    # Add RDKit features to result
+    result = pd.concat([result, rdkit_features_df], axis=1)
+
+    # Compute Mordred descriptors
+    if include_mordred:
+        logger.info("Computing Mordred descriptors from relevant modules...")
+        calc = MordredCalculator()
+
+        # Register 5 ADMET-focused modules (avoiding overlap with RDKit)
+        calc.register(AcidBase)  # ~2 descriptors: nAcid, nBase
+        calc.register(Aromatic)  # ~2 descriptors: nAromAtom, nAromBond
+        calc.register(Constitutional)  # ~30 descriptors: structural complexity
+        calc.register(Chi)  # ~32 descriptors: connectivity indices
+        calc.register(CarbonTypes)  # ~20 descriptors: carbon hybridization
+
+        # Compute Mordred descriptors
+        valid_mols = [mol if mol is not None else Chem.MolFromSmiles("C") for mol in molecules]
+        mordred_df = calc.pandas(valid_mols, nproc=1)  # For serverless, use nproc=1
+
+        # Replace values for invalid molecules with NaN
+        for i, mol in enumerate(molecules):
+            if mol is None:
+                mordred_df.iloc[i] = np.nan
+
+        # Handle Mordred's special error values
+        for col in mordred_df.columns:
+            mordred_df[col] = pd.to_numeric(mordred_df[col], errors="coerce")
+
+        # Set index to match result DataFrame
+        mordred_df.index = result.index
+
+        # Add Mordred features to result
+        result = pd.concat([result, mordred_df], axis=1)
+
+    # Compute stereochemistry features if requested
+    if include_stereo:
+        logger.info("Computing Stereochemistry Descriptors...")
+
+        stereo_features = []
+        for mol in molecules:
+            stereo_dict = compute_stereochemistry_features(mol)
+            stereo_features.append(stereo_dict)
+
+        # Create stereochemistry DataFrame
+        stereo_df = pd.DataFrame(stereo_features, index=result.index)
+
+        # Add stereochemistry features to result
+        result = pd.concat([result, stereo_df], axis=1)
+
+        logger.info(f"Added {len(stereo_df.columns)} stereochemistry descriptors")
+
+    # Log summary
+    valid_mols = sum(1 for m in molecules if m is not None)
+    total_descriptors = len(result.columns) - len(df.columns)
+    logger.info(f"Computed {total_descriptors} descriptors for {valid_mols}/{len(df)} valid molecules")
+
+    # Log descriptor breakdown
+    rdkit_count = len(rdkit_features_df.columns)
+    mordred_count = len(mordred_df.columns) if include_mordred else 0
+    stereo_count = len(stereo_df.columns) if include_stereo else 0
+    logger.info(f"Descriptor breakdown: RDKit={rdkit_count}, Mordred={mordred_count}, Stereo={stereo_count}")
+
+    # Sanitize column names for AWS Athena compatibility
+    # - Must be lowercase, no special characters except underscore, no spaces
+    result.columns = [re.sub(r"_+", "_", re.sub(r"[^a-z0-9_]", "_", col.lower())) for col in result.columns]
+
+    # Drop duplicate columns if any exist after sanitization
+    if result.columns.duplicated().any():
+        logger.warning("Duplicate column names after sanitization - dropping duplicates!")
+        result = result.loc[:, ~result.columns.duplicated()]
+
+    return result
+
+
+if __name__ == "__main__":
+    import time
+    from mol_standardize import standardize
+    from workbench.api import DataSource
+
+    # Configure pandas display
+    pd.set_option("display.max_columns", None)
+    pd.set_option("display.max_colwidth", 100)
+    pd.set_option("display.width", 1200)
+
+    # Test data - stereochemistry examples
+    stereo_test_data = pd.DataFrame(
+        {
+            "smiles": [
+                "CC(=O)Oc1ccccc1C(=O)O",  # Aspirin
+                "C[C@H](N)C(=O)O",  # L-Alanine
+                "C[C@@H](N)C(=O)O",  # D-Alanine
+                "C/C=C/C=C/C",  # E,E-hexadiene
+                "CC(F)(Cl)Br",  # Unspecified chiral
+                "",
+                "INVALID",  # Invalid cases
+            ],
+            "name": ["Aspirin", "L-Alanine", "D-Alanine", "E,E-hexadiene", "Unspecified", "Empty", "Invalid"],
+        }
+    )
+
+    # Test data - salt handling examples
+    salt_test_data = pd.DataFrame(
+        {
+            "smiles": [
+                "CC(=O)O",  # Acetic acid
+                "[Na+].CC(=O)[O-]",  # Sodium acetate
+                "CC(C)NCC(O)c1ccc(O)c(O)c1.Cl",  # Drug HCl salt
+                "Oc1ccccn1",  # Tautomer 1
+                "O=c1cccc[nH]1",  # Tautomer 2
+            ],
+            "compound_id": [f"C{i:03d}" for i in range(1, 6)],
+        }
+    )
+
+    def run_basic_tests():
+        """Run basic functionality tests"""
+        print("=" * 80)
+        print("BASIC FUNCTIONALITY TESTS")
+        print("=" * 80)
+
+        # Test stereochemistry
+        result = compute_descriptors(stereo_test_data, include_stereo=True)
+
+        print("\nStereochemistry features (selected molecules):")
+        for idx, name in enumerate(stereo_test_data["name"][:4]):
+            print(
+                f"{name:15} - centers: {result.iloc[idx]['num_stereocenters']:.0f}, "
+                f"R/S: {result.iloc[idx]['num_r_centers']:.0f}/"
+                f"{result.iloc[idx]['num_s_centers']:.0f}"
+            )
+
+        # Test salt handling
+        print("\nSalt extraction test:")
+        std_result = standardize(salt_test_data, extract_salts=True)
+        for _, row in std_result.iterrows():
+            salt_info = f" → salt: {row['salt']}" if pd.notna(row["salt"]) else ""
+            print(f"{row['compound_id']}: {row['smiles'][:30]}{salt_info}")
+
+    def run_performance_tests():
+        """Run performance timing tests"""
+        print("\n" + "=" * 80)
+        print("PERFORMANCE TESTS on real world molecules")
+        print("=" * 80)
+
+        # Get a real dataset from Workbench
+        ds = DataSource("aqsol_data")
+        df = ds.pull_dataframe()[["id", "smiles"]][:1000]  # Limit to 1000 for testing
+        n_mols = df.shape[0]
+        print(f"Pulled {n_mols} molecules from DataSource 'aqsol_data'")
+
+        # Test configurations
+        configs = [
+            ("Standardize (full)", standardize, {"extract_salts": True, "canonicalize_tautomer": True}),
+            ("Standardize (minimal)", standardize, {"extract_salts": False, "canonicalize_tautomer": False}),
+            ("Descriptors (all)", compute_descriptors, {"include_mordred": True, "include_stereo": True}),
+            ("Descriptors (RDKit only)", compute_descriptors, {"include_mordred": False, "include_stereo": False}),
+        ]
+
+        results = []
+        for name, func, params in configs:
+            start = time.time()
+            _ = func(df, **params)
+            elapsed = time.time() - start
+            throughput = n_mols / elapsed
+            results.append((name, elapsed, throughput))
+            print(f"{name:25} {elapsed:6.2f}s ({throughput:6.1f} mol/s)")
+
+        # Full pipeline test
+        print("\nFull pipeline (standardize + all descriptors):")
+        start = time.time()
+        std_data = standardize(df)
+        standardize_time = time.time() - start
+        print(f"  Standardize: {standardize_time:.2f}s ({n_mols / standardize_time:.1f} mol/s)")
+        start = time.time()
+        _ = compute_descriptors(std_data)
+        descriptor_time = time.time() - start
+        print(f"  Descriptors: {descriptor_time:.2f}s ({n_mols / descriptor_time:.1f} mol/s)")
+        pipeline_time = standardize_time + descriptor_time
+        print(f"  Total: {pipeline_time:.2f}s ({n_mols / pipeline_time:.1f} mol/s)")
+
+        return results
+
+    # Run tests
+    run_basic_tests()
+    timing_results = run_performance_tests()
+
+    print("\n✅ All tests completed!")