M3Drop 0.4.42__py3-none-any.whl → 0.4.45__py3-none-any.whl

m3Drop/DiagnosticsCPU.py

@@ -0,0 +1,401 @@
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+import h5py
+import os
+import time
+import pickle
+import gc
+from scipy import sparse
+from scipy import stats
+import anndata
+
+import statsmodels.api as sm
+from scipy.stats import norm
+from statsmodels.stats.multitest import multipletests
+
+# [FIX] Strict Relative Imports
+from .ControlDeviceCPU import ControlDevice
+from .CoreCPU import hidden_calc_valsCPU, NBumiFitModelCPU, NBumiFitDispVsMeanCPU, dropout_prob_kernel_cpu
+
+# ==========================================
+#        DIAGNOSTICS & COMPARISON (CPU)
+# ==========================================
+
+def NBumiFitBasicModelCPU(
+    filename: str,
+    stats: dict,
+    mask_filename: str = None, 
+    mode: str = "auto",
+    manual_target: int = 3000,
+    phase_label: str = "Phase [1/1]",
+    desc_label: str = None
+) -> dict:
+    """
+    Fits the Basic Model by calculating Normalized Variance ON-THE-FLY (CPU Optimized).
+    STRICT FLOAT64 ENFORCEMENT.
+    """
+    # 1. Get Raw Dimensions & Setup ControlDevice
+    with h5py.File(filename, 'r') as f:
+        indptr_cpu = f['X']['indptr'][:]
+        total_rows = len(indptr_cpu) - 1
+        raw_ng = f['X'].attrs['shape'][1] 
+
+    device = ControlDevice(
+        indptr=indptr_cpu, 
+        total_rows=total_rows, 
+        n_genes=raw_ng, 
+        mode=mode, 
+        manual_target=manual_target
+    )
+    nc = device.total_rows
+
+    if desc_label:
+        print(f"{phase_label}: {desc_label}")
+
+    # 2. Load Mask
+    if mask_filename and os.path.exists(mask_filename):
+        with open(mask_filename, 'rb') as f:
+            mask = pickle.load(f)
+    else:
+        mask = np.ones(raw_ng, dtype=bool)
+
+    filtered_ng = int(np.sum(mask))
+    
+    # 3. Pre-calculate Size Factors
+    cell_sums = stats['tis'].values
+    median_sum = np.median(cell_sums[cell_sums > 0])
+    
+    # [FLOAT64] Explicit enforcement
+    size_factors = np.ones_like(cell_sums, dtype=np.float64)
+    non_zero_mask = cell_sums > 0
+    size_factors[non_zero_mask] = cell_sums[non_zero_mask] / median_sum
+    
+    # 4. Init Accumulators
+    sum_norm_x = np.zeros(filtered_ng, dtype=np.float64)
+    sum_norm_sq = np.zeros(filtered_ng, dtype=np.float64)
+
+    with h5py.File(filename, 'r') as f_in:
+        h5_indptr = f_in['X']['indptr']
+        h5_data = f_in['X']['data']
+        h5_indices = f_in['X']['indices']
+
+        current_row = 0
+        while current_row < nc:
+            # CPU prefers dense chunks if they fit in L3, but sparse is safer for RAM.
+            # We use 'dense' mode here because we convert to dense for normalization anyway.
+            end_row = device.get_next_chunk(current_row, mode='dense', overhead_multiplier=1.5)
+            if end_row is None or end_row <= current_row: break
+
+            chunk_size = end_row - current_row
+            print(f"{phase_label}: Processing {end_row} of {nc} | Chunk: {chunk_size}", end='\r')
+
+            start_idx, end_idx = h5_indptr[current_row], h5_indptr[end_row]
+            if start_idx == end_idx:
+                current_row = end_row
+                continue
+            
+            # [FLOAT64] Load Raw Chunk
+            data = np.array(h5_data[start_idx:end_idx], dtype=np.float64)
+            indices = np.array(h5_indices[start_idx:end_idx])
+            indptr = np.array(h5_indptr[current_row:end_row+1] - h5_indptr[current_row])
+            
+            # Reconstruct CSR & Filter
+            raw_chunk = sparse.csr_matrix((data, indices, indptr), shape=(chunk_size, raw_ng))
+            filtered_chunk = raw_chunk[:, mask]
+
+            # Normalization (Vectorized CPU)
+            sf_chunk = size_factors[current_row:end_row]
+            
+            # Scipy sparse multiplication is efficient
+            # D = diag(1/sf)
+            recip_sf = 1.0 / sf_chunk
+            D = sparse.diags(recip_sf)
+            norm_chunk = D.dot(filtered_chunk)
+            
+            # Rounding (in-place on data array)
+            np.round(norm_chunk.data, out=norm_chunk.data)
+            
+            # Accumulate
+            # Convert to dense for summation if chunk is small (faster on CPU)
+            # or keep sparse if very large. Given L3 optimization, dense is often fine.
+            norm_dense = norm_chunk.toarray() 
+            
+            sum_norm_x += norm_dense.sum(axis=0)
+            sum_norm_sq += (norm_dense ** 2).sum(axis=0)
+            
+            current_row = end_row
+
+    # Final Calculations
+    mean_norm = sum_norm_x / nc
+    mean_sq_norm = sum_norm_sq / nc
+    var_norm = mean_sq_norm - (mean_norm ** 2)
+    
+    denom = var_norm - mean_norm
+    sizes = np.full(filtered_ng, 1000.0, dtype=np.float64)
+    valid_mask = denom > 1e-6
+    sizes[valid_mask] = mean_norm[valid_mask]**2 / denom[valid_mask]
+    
+    # Filtering outliers (Numpy version)
+    with np.errstate(invalid='ignore'):
+        max_size_val = np.nanmax(sizes[sizes < 1e6]) * 10
+        
+    if np.isnan(max_size_val) or max_size_val == 0: max_size_val = 1000.0
+    sizes[np.isnan(sizes) | (sizes <= 0)] = max_size_val
+    sizes[sizes < 1e-10] = 1e-10
+    
+    print("") 
+    print(f"{phase_label}: COMPLETE")
+
+    return {
+        'var_obs': pd.Series(var_norm, index=stats['tjs'].index),
+        'sizes': pd.Series(sizes, index=stats['tjs'].index),
+        'vals': stats 
+    }
+
+def NBumiCheckFitFSCPU(
+    filename: str,
+    fit: dict,
+    mode: str = "auto",
+    manual_target: int = 3000,
+    suppress_plot=False,
+    plot_filename=None,
+    phase_label="Phase [1/1]",
+    desc_label: str = None
+) -> dict:
+    """
+    Calculates expected dropouts using NUMBA KERNEL on CPU.
+    """
+    vals = fit['vals']
+    ng = vals['ng']
+    
+    with h5py.File(filename, 'r') as f:
+        indptr_cpu = f['X']['indptr'][:]
+        total_rows = len(indptr_cpu) - 1
+    
+    device = ControlDevice(
+        indptr=indptr_cpu, 
+        total_rows=total_rows, 
+        n_genes=ng, 
+        mode=mode, 
+        manual_target=manual_target
+    )
+    nc = device.total_rows
+
+    if desc_label:
+        print(f"{phase_label}: {desc_label}")
+
+    size_coeffs = NBumiFitDispVsMeanCPU(fit, suppress_plot=True)
+    
+    tjs = vals['tjs'].values.astype(np.float64)
+    tis = vals['tis'].values.astype(np.float64)
+    total = vals['total']
+
+    mean_expression = tjs / nc
+    log_mean_expression = np.zeros_like(mean_expression)
+    valid_means = mean_expression > 0
+    log_mean_expression[valid_means] = np.log(mean_expression[valid_means])
+    smoothed_size = np.exp(size_coeffs[0] + size_coeffs[1] * log_mean_expression)
+
+    row_ps = np.zeros(ng, dtype=np.float64)
+    col_ps = np.zeros(nc, dtype=np.float64)
+
+    current_row = 0
+    while current_row < nc:
+        # Use dense mode for Numba efficiency
+        end_row = device.get_next_chunk(current_row, mode='dense', overhead_multiplier=1.1)
+        if end_row is None or end_row <= current_row: break
+        
+        chunk_size = end_row - current_row
+        print(f"{phase_label}: Processing {end_row} of {nc} | Chunk: {chunk_size}", end='\r')
+
+        tis_chunk = tis[current_row:end_row]
+        
+        # [CRITICAL] NUMBA KERNEL CALL
+        # Prepare output buffer
+        p_is_chunk = np.empty((chunk_size, ng), dtype=np.float64)
+        
+        dropout_prob_kernel_cpu(
+            tjs,                 # Gene totals
+            tis_chunk,           # Cell totals (1D array, broadcasting handled inside kernel)
+            total,               # Grand total
+            smoothed_size,       # Exp size
+            p_is_chunk           # Output buffer
+        )
+
+        # Sanitize
+        p_is_chunk = np.nan_to_num(p_is_chunk, nan=0.0, posinf=1.0, neginf=0.0)
+        
+        row_ps += p_is_chunk.sum(axis=0)
+        col_ps[current_row:end_row] = p_is_chunk.sum(axis=1)
+        
+        current_row = end_row
+
+    print("")
+    print(f"{phase_label}: COMPLETE")
+
+    return {
+        'rowPs': pd.Series(row_ps, index=fit['vals']['tjs'].index),
+        'colPs': pd.Series(col_ps, index=fit['vals']['tis'].index)
+    }
+
+def NBumiCompareModelsCPU(
+    raw_filename: str,
+    stats: dict,
+    fit_adjust: dict,
+    mask_filename: str = None, 
+    mode: str = "auto",
+    manual_target: int = 3000,
+    suppress_plot=False,
+    plot_filename=None
+) -> dict:
+    print(f"FUNCTION: NBumiCompareModelsCPU()")
+    pipeline_start_time = time.time()
+
+    # STEP 1: Fit Basic Model
+    fit_basic = NBumiFitBasicModelCPU(
+        raw_filename, 
+        stats, 
+        mask_filename=mask_filename,
+        mode=mode, 
+        manual_target=manual_target,
+        phase_label="Phase [1/3]",
+        desc_label="Fitting Basic Model (Virtual)..."
+    )
+    
+    # STEP 2: Depth-Adjusted Dropout
+    check_adjust = NBumiCheckFitFSCPU(
+        raw_filename, 
+        fit_adjust, 
+        mode=mode, 
+        manual_target=manual_target, 
+        suppress_plot=True,
+        phase_label="Phase [2/3]",
+        desc_label="Calculating Depth-Adjusted Dropouts..."
+    )
+    
+    # STEP 3: Basic Dropout
+    stats_virtual = stats.copy()
+    mean_depth = stats['total'] / stats['nc']
+    stats_virtual['tis'] = pd.Series(
+        np.full(stats['nc'], mean_depth), 
+        index=stats['tis'].index
+    )
+    
+    fit_basic_for_eval = {
+        'sizes': fit_basic['sizes'],
+        'vals': stats_virtual,
+        'var_obs': fit_basic['var_obs']
+    }
+    
+    check_basic = NBumiCheckFitFSCPU(
+        raw_filename, 
+        fit_basic_for_eval, 
+        mode=mode, 
+        manual_target=manual_target, 
+        suppress_plot=True,
+        phase_label="Phase [3/3]",
+        desc_label="Calculating Basic Dropouts..."
+    )
+
+    # Calculation & Plotting
+    nc_data = stats['nc']
+    mean_expr = stats['tjs'] / nc_data
+    observed_dropout = stats['djs'] / nc_data
+    
+    adj_dropout_fit = check_adjust['rowPs'] / nc_data
+    bas_dropout_fit = check_basic['rowPs'] / nc_data
+    
+    err_adj = np.sum(np.abs(adj_dropout_fit - observed_dropout))
+    err_bas = np.sum(np.abs(bas_dropout_fit - observed_dropout))
+    
+    comparison_df = pd.DataFrame({
+        'mean_expr': mean_expr,
+        'observed': observed_dropout,
+        'adj_fit': adj_dropout_fit,
+        'bas_fit': bas_dropout_fit
+    })
+    
+    # Plotting Logic (Standard Matplotlib)
+    plt.figure(figsize=(10, 6))
+    sorted_idx = np.argsort(mean_expr.values)
+    plot_idx = sorted_idx[::2] if len(mean_expr) > 20000 else sorted_idx
+
+    plt.scatter(mean_expr.iloc[plot_idx], observed_dropout.iloc[plot_idx], 
+                c='black', s=3, alpha=0.5, label='Observed')
+    
+    plt.scatter(mean_expr.iloc[plot_idx], bas_dropout_fit.iloc[plot_idx], 
+                c='purple', s=3, alpha=0.6, label=f'Basic Fit (Error: {err_bas:.2f})')
+    
+    plt.scatter(mean_expr.iloc[plot_idx], adj_dropout_fit.iloc[plot_idx], 
+                c='goldenrod', s=3, alpha=0.7, label=f'Depth-Adjusted Fit (Error: {err_adj:.2f})')
+    
+    plt.xscale('log')
+    plt.xlabel("Mean Expression")
+    plt.ylabel("Dropout Rate")
+    plt.title("M3Drop Model Comparison (CPU)")
+    plt.legend()
+    plt.grid(True, linestyle='--', alpha=0.3)
+
+    if plot_filename:
+        plt.savefig(plot_filename, dpi=300, bbox_inches='tight')
+        print(f"Saving plot to: {plot_filename}")
+
+    if not suppress_plot:
+        plt.show()
+    
+    plt.close()
+
+    pipeline_end_time = time.time()
+    print(f"Total time: {pipeline_end_time - pipeline_start_time:.2f} seconds.\n")
+    
+    return {
+        "errors": {"Depth-Adjusted": err_adj, "Basic": err_bas},
+        "comparison_df": comparison_df
+    }
+
+def NBumiPlotDispVsMeanCPU(
+    fit: dict,
+    suppress_plot: bool = False,
+    plot_filename: str = None
+):
+    print("FUNCTION: NBumiPlotDispVsMeanCPU()")
+    start_time = time.time()
+        
+    mean_expression = fit['vals']['tjs'].values / fit['vals']['nc']
+    sizes = fit['sizes'].values
+    
+    coeffs = NBumiFitDispVsMeanCPU(fit, suppress_plot=True)
+    intercept, slope = coeffs[0], coeffs[1]
+
+    log_mean_expr_range = np.linspace(
+        np.log(mean_expression[mean_expression > 0].min()),
+        np.log(mean_expression.max()),
+        100
+    )
+    log_fitted_sizes = intercept + slope * log_mean_expr_range
+    fitted_sizes = np.exp(log_fitted_sizes)
+
+    plt.figure(figsize=(8, 6))
+    plt.scatter(mean_expression, sizes, label='Observed Dispersion', alpha=0.5, s=8)
+    plt.plot(np.exp(log_mean_expr_range), fitted_sizes, color='red', label='Regression Fit', linewidth=2)
+
+    plt.xscale('log')
+    plt.yscale('log')
+    plt.xlabel('Mean Expression')
+    plt.ylabel('Dispersion Parameter (Sizes)')
+    plt.title('Dispersion vs. Mean Expression (CPU)')
+    plt.legend()
+    plt.grid(True, which="both", linestyle='--', alpha=0.6)
+
+    if plot_filename:
+        plt.savefig(plot_filename, dpi=300, bbox_inches='tight')
+        print(f"Saving plot to: {plot_filename}")
+            
+    if not suppress_plot:
+        plt.show()
+
+    plt.close()
+    
+    end_time = time.time()
+    print(f"Total time: {end_time - start_time:.2f} seconds.\n")