M3Drop 0.4.58__tar.gz → 0.4.60__tar.gz

{m3drop-0.4.58 → m3drop-0.4.60/M3Drop.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: M3Drop
-Version: 0.4.58
+Version: 0.4.60
 Summary: A Python implementation of the M3Drop single-cell RNA-seq analysis tool.
 Home-page: https://github.com/PragalvhaSharma/m3DropNew
 Author: Tallulah Andrews

{m3drop-0.4.58/M3Drop.egg-info → m3drop-0.4.60}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: M3Drop
-Version: 0.4.58
+Version: 0.4.60
 Summary: A Python implementation of the M3Drop single-cell RNA-seq analysis tool.
 Home-page: https://github.com/PragalvhaSharma/m3DropNew
 Author: Tallulah Andrews

{m3drop-0.4.58 → m3drop-0.4.60}/m3Drop/CoreCPU.py

@@ -22,10 +22,19 @@ from scipy.stats import norm
 from scipy import sparse
 from statsmodels.stats.multitest import multipletests
 
-# [FIX] Strict Relative Import
-# This ensures that if ControlDeviceCPU fails to load (e.g. missing dependency), 
-# the real error is shown instead of being masked.
-from .ControlDeviceCPU import ControlDevice
+# ==========================================
+#        HYBRID IMPORT (PACKAGE VS LOCAL)
+# ==========================================
+try:
+    # Case 1: Running as an installed package
+    from .ControlDeviceCPU import ControlDevice
+except ImportError:
+    # Case 2: Running locally
+    try:
+        from ControlDeviceCPU import ControlDevice
+    except ImportError:
+        print("CRITICAL ERROR: 'ControlDeviceCPU.py' not found.")
+        sys.exit(1)
 
 # ==========================================
 #        NUMBA KERNELS (CPU OPTIMIZED)
@@ -323,9 +332,17 @@ def NBumiFitDispVsMeanCPU(fit: dict, suppress_plot=True):
     tjs = vals['tjs'].values
     mean_expression = tjs / vals['nc']
     
-    forfit = (np.isfinite(size_g)) & (size_g < 1e6) & (mean_expression > 1e-3) & (size_g > 0)
+    # [FIX] Filter out the 10,000 imputation values.
+    # We treat 10,000 as an error code for "Under-Dispersed/Poissonian".
+    # Masking these prevents artificial gravity from pulling the regression line up.
+    forfit = (np.isfinite(size_g)) & \
+             (size_g < 9999.0) & \
+             (mean_expression > 1e-3) & \
+             (size_g > 0)
+
     log2_mean_expr = np.log2(mean_expression, where=(mean_expression > 0))
     
+    # Heuristic: If we have enough high-expression genes, focus fit there
     higher = log2_mean_expr > 4
     if np.sum(higher & forfit) > 2000:
         forfit = higher & forfit
@@ -338,8 +355,10 @@ def NBumiFitDispVsMeanCPU(fit: dict, suppress_plot=True):
     
     if not suppress_plot:
         plt.figure(figsize=(7, 6))
-        plt.scatter(x, y, alpha=0.5, s=1)
-        plt.plot(x, model.fittedvalues, color='red')
+        # Plot ALL points (grey) vs FITTED points (blue)
+        plt.scatter(np.log(mean_expression), np.log(size_g), alpha=0.3, s=1, c='grey', label='All Genes')
+        plt.scatter(x, y, alpha=0.5, s=1, c='blue', label='Used for Fit')
+        plt.plot(x, model.fittedvalues, color='red', label='Regression Fit')
         plt.show()
 
     return model.params

{m3drop-0.4.58 → m3drop-0.4.60}/m3Drop/CoreGPU.py

@@ -22,8 +22,19 @@ import matplotlib.pyplot as plt
 from scipy.stats import norm
 from statsmodels.stats.multitest import multipletests
 
-# Package-compatible import
-from .ControlDeviceGPU import ControlDevice
+# ==========================================
+#      HYBRID IMPORT (PACKAGE VS LOCAL)
+# ==========================================
+try:
+    # Case 1: Running as an installed package (e.g. import m3drop.CoreGPU)
+    from .ControlDeviceGPU import ControlDevice
+except ImportError:
+    # Case 2: Running locally (e.g. python CoreGPU.py)
+    try:
+        from ControlDeviceGPU import ControlDevice
+    except ImportError:
+        print("CRITICAL ERROR: 'ControlDeviceGPU.py' not found.")
+        sys.exit(1)
 
 # ==========================================
 #        FUSED KERNELS
@@ -313,9 +324,14 @@ def NBumiFitDispVsMeanGPU(fit: dict, suppress_plot=True):
     tjs = vals['tjs'].values
     mean_expression = tjs / vals['nc']
     
-    forfit = (np.isfinite(size_g)) & (size_g < 1e6) & (mean_expression > 1e-3) & (size_g > 0)
+    forfit = (np.isfinite(size_g)) & \
+             (size_g < 9999.0) & \
+             (mean_expression > 1e-3) & \
+             (size_g > 0)
+    
     log2_mean_expr = np.log2(mean_expression, where=(mean_expression > 0))
     
+    # Heuristic: If we have enough high-expression genes, focus fit there
     higher = log2_mean_expr > 4
     if np.sum(higher & forfit) > 2000:
         forfit = higher & forfit
@@ -328,7 +344,8 @@ def NBumiFitDispVsMeanGPU(fit: dict, suppress_plot=True):
     
     if not suppress_plot:
         plt.figure(figsize=(7, 6))
-        plt.scatter(x, y, alpha=0.5, s=1)
+        # Visual check only - code below handles the production plot
+        plt.scatter(x, y, alpha=0.5, s=1, c='blue')
         plt.plot(x, model.fittedvalues, color='red')
         plt.show()
 

{m3drop-0.4.58 → m3drop-0.4.60}/m3Drop/DiagnosticsCPU.py

@@ -4,6 +4,7 @@ import matplotlib.pyplot as plt
 import h5py
 import os
 import time
+import sys
 import pickle
 import gc
 from scipy import sparse
@@ -14,9 +15,29 @@ 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
+# ==========================================
+#        HYBRID IMPORT (PACKAGE VS LOCAL)
+# ==========================================
+
+# [FIX] Hybrid Import: ControlDeviceCPU
+try:
+    from .ControlDeviceCPU import ControlDevice
+except ImportError:
+    try:
+        from ControlDeviceCPU import ControlDevice
+    except ImportError:
+        print("CRITICAL ERROR: 'ControlDeviceCPU.py' not found.")
+        sys.exit(1)
+
+# [FIX] Hybrid Import: CoreCPU
+try:
+    from .CoreCPU import hidden_calc_valsCPU, NBumiFitModelCPU, NBumiFitDispVsMeanCPU, dropout_prob_kernel_cpu
+except ImportError:
+    try:
+        from CoreCPU import hidden_calc_valsCPU, NBumiFitModelCPU, NBumiFitDispVsMeanCPU, dropout_prob_kernel_cpu
+    except ImportError:
+        print("CRITICAL ERROR: 'CoreCPU.py' not found.")
+        sys.exit(1)
 
 # ==========================================
 #        DIAGNOSTICS & COMPARISON (CPU)
@@ -244,7 +265,7 @@ def NBumiCompareModelsCPU(
     stats: dict,
     fit_adjust: dict,
     mask_filename: str = None, 
-    mode: str = "auto",
+    mode: str = "auto", 
     manual_target: int = 3000,
     suppress_plot=False,
     plot_filename=None
@@ -365,9 +386,11 @@ def NBumiPlotDispVsMeanCPU(
     mean_expression = fit['vals']['tjs'].values / fit['vals']['nc']
     sizes = fit['sizes'].values
     
+    # 1. Get calibrated coefficients (using the fixed CoreCPU logic)
     coeffs = NBumiFitDispVsMeanCPU(fit, suppress_plot=True)
     intercept, slope = coeffs[0], coeffs[1]
 
+    # 2. Calculate the regression line
     log_mean_expr_range = np.linspace(
         np.log(mean_expression[mean_expression > 0].min()),
         np.log(mean_expression.max()),
@@ -376,8 +399,19 @@ def NBumiPlotDispVsMeanCPU(
     log_fitted_sizes = intercept + slope * log_mean_expr_range
     fitted_sizes = np.exp(log_fitted_sizes)
 
+    # 3. [FIX] Mask the 10k outliers for the SCATTER PLOT
+    # We create a visualization mask to hide the distracting "roof" at y=10000
+    mask_viz = (sizes < 9999.0) & (sizes > 0)
+    
+    mean_expr_clean = mean_expression[mask_viz]
+    sizes_clean = sizes[mask_viz]
+
     plt.figure(figsize=(8, 6))
-    plt.scatter(mean_expression, sizes, label='Observed Dispersion', alpha=0.5, s=8)
+    
+    # Plot only the clean data
+    plt.scatter(mean_expr_clean, sizes_clean, label='Observed Dispersion', alpha=0.5, s=8)
+    
+    # Plot the regression line (calculated correctly via CoreCPU)
     plt.plot(np.exp(log_mean_expr_range), fitted_sizes, color='red', label='Regression Fit', linewidth=2)
 
     plt.xscale('log')

{m3drop-0.4.58 → m3drop-0.4.60}/m3Drop/DiagnosticsGPU.py

@@ -9,17 +9,36 @@ import time
 import pickle
 import psutil
 import gc
+import sys
 from scipy import sparse
 from scipy import stats
 import anndata
 
-from .ControlDeviceGPU import ControlDevice
-from .CoreGPU import (
-    hidden_calc_valsGPU, 
-    NBumiFitModelGPU, 
-    NBumiFitDispVsMeanGPU, 
-    dropout_prob_kernel
-)
+# ==========================================
+#      HYBRID IMPORT (PACKAGE VS LOCAL)
+# ==========================================
+try:
+    # Case 1: Package
+    from .ControlDeviceGPU import ControlDevice
+    from .CoreGPU import (
+        hidden_calc_valsGPU, 
+        NBumiFitModelGPU, 
+        NBumiFitDispVsMeanGPU, 
+        dropout_prob_kernel
+    )
+except ImportError:
+    # Case 2: Local
+    try:
+        from ControlDeviceGPU import ControlDevice
+        from CoreGPU import (
+            hidden_calc_valsGPU, 
+            NBumiFitModelGPU, 
+            NBumiFitDispVsMeanGPU, 
+            dropout_prob_kernel
+        )
+    except ImportError:
+        print("CRITICAL ERROR: Dependencies (ControlDeviceGPU, CoreGPU) not found.")
+        sys.exit(1)
 
 from cupy.sparse import csr_matrix as cp_csr_matrix
 import scipy.sparse as sp
@@ -385,9 +404,11 @@ def NBumiPlotDispVsMeanGPU(
     mean_expression = fit['vals']['tjs'].values / fit['vals']['nc']
     sizes = fit['sizes'].values
     
+    # 1. Get calibrated coefficients (using the fixed CoreGPU logic)
     coeffs = NBumiFitDispVsMeanGPU(fit, suppress_plot=True)
     intercept, slope = coeffs[0], coeffs[1]
 
+    # 2. Calculate the regression line
     log_mean_expr_range = np.linspace(
         np.log(mean_expression[mean_expression > 0].min()),
         np.log(mean_expression.max()),
@@ -396,8 +417,19 @@ def NBumiPlotDispVsMeanGPU(
     log_fitted_sizes = intercept + slope * log_mean_expr_range
     fitted_sizes = np.exp(log_fitted_sizes)
 
+    # 3. [FIX] Mask the 10k outliers for the SCATTER PLOT
+    # We create a visualization mask to hide the distracting "roof" at y=10000
+    mask_viz = (sizes < 9999.0) & (sizes > 0)
+    
+    mean_expr_clean = mean_expression[mask_viz]
+    sizes_clean = sizes[mask_viz]
+
     plt.figure(figsize=(8, 6))
-    plt.scatter(mean_expression, sizes, label='Observed Dispersion', alpha=0.5, s=8)
+    
+    # Plot only the clean data
+    plt.scatter(mean_expr_clean, sizes_clean, label='Observed Dispersion', alpha=0.5, s=8)
+    
+    # Plot the regression line (which was calculated correctly via CoreGPU)
     plt.plot(np.exp(log_mean_expr_range), fitted_sizes, color='red', label='Regression Fit', linewidth=2)
 
     plt.xscale('log')

{m3drop-0.4.58 → m3drop-0.4.60}/m3Drop/NormalizationCPU.py

@@ -16,8 +16,19 @@ except ImportError:
     print("CRITICAL ERROR: 'numba' not found. Please install it (pip install numba).")
     sys.exit(1)
 
-# Strict Relative Import
-from .ControlDeviceCPU import ControlDevice
+# ==========================================
+#        HYBRID IMPORT (PACKAGE VS LOCAL)
+# ==========================================
+try:
+    # Case 1: Running as an installed package
+    from .ControlDeviceCPU import ControlDevice
+except ImportError:
+    # Case 2: Running locally
+    try:
+        from ControlDeviceCPU import ControlDevice
+    except ImportError:
+        print("CRITICAL ERROR: 'ControlDeviceCPU.py' not found.")
+        sys.exit(1)
 
 # ==========================================
 #        NUMBA KERNELS (CPU)

{m3drop-0.4.58 → m3drop-0.4.60}/m3Drop/NormalizationGPU.py

@@ -18,10 +18,14 @@ except ImportError:
     cupy = None
     HAS_GPU = False
 
-# Package-compatible import
+# ==========================================
+#      HYBRID IMPORT (PACKAGE VS LOCAL)
+# ==========================================
 try:
+    # Case 1: Package
     from .ControlDeviceGPU import ControlDevice
 except ImportError:
+    # Case 2: Local
     try:
         from ControlDeviceGPU import ControlDevice
     except ImportError:
@@ -367,6 +371,3 @@ def NBumiPearsonResidualsCombinedGPU(
 
     if hasattr(adata_in, "file") and adata_in.file is not None: adata_in.file.close()
     print(f"Total time: {time.perf_counter() - start_time:.2f} seconds.\n")
-
-
-

{m3drop-0.4.58 → m3drop-0.4.60}/setup.py

@@ -5,7 +5,7 @@ with open("README.md", "r", encoding="utf-8") as fh:
 
 setuptools.setup(
     name="M3Drop",  # Name for pip (pip install M3Drop)
-    version="0.4.58",
+    version="0.4.60",
     author="Tallulah Andrews",
     author_email="tandrew6@uwo.ca",
     description="A Python implementation of the M3Drop single-cell RNA-seq analysis tool.",