mcDETECT 2.0.3__py3-none-any.whl → 2.0.5__py3-none-any.whl

mcDETECT/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "2.0.3"
+__version__ = "2.0.5"
 
 from . import model
 from . import utils

mcDETECT/model.py

@@ -4,7 +4,9 @@ import miniball
 import numpy as np
 import pandas as pd
 import scanpy as sc
+from collections import Counter
 from rtree import index
+from scipy.sparse import csr_matrix
 from scipy.spatial import cKDTree
 from scipy.stats import poisson
 from shapely.geometry import Point
@@ -18,12 +20,12 @@ from .utils import *
 class mcDETECT:
     
     
-    def __init__(self, type, transcripts, syn_genes, nc_genes = None, eps = 1.5, minspl = None, grid_len = 1.0, cutoff_prob = 0.95, alpha = 5.0, low_bound = 3,
+    def __init__(self, type, transcripts, gnl_genes, nc_genes = None, eps = 1.5, minspl = None, grid_len = 1.0, cutoff_prob = 0.95, alpha = 5.0, low_bound = 3,
                  size_thr = 4.0, in_nucleus_thr = (0.5, 0.5), l = 1.0, rho = 0.2, s = 1.0, nc_top = 20, nc_thr = 0.1):
         
         self.type = type                        # string, iST platform, now support MERSCOPE, Xenium, and CosMx
         self.transcripts = transcripts          # dataframe, transcripts file
-        self.syn_genes = syn_genes              # list, string, all synaptic markers
+        self.gnl_genes = gnl_genes              # list, string, all granule markers
         self.nc_genes = nc_genes                # list, string, all negative controls
         self.eps = eps                          # numeric, searching radius epsilon
         self.minspl = minspl                    # integer, manually select min_samples, i.e., no automatic parameter selection
@@ -57,10 +59,11 @@ class mcDETECT:
     
     # [INNER] calculate tissue area, input for poisson_select()
     def tissue_area(self):
-        x_bins, y_bins = self.construct_grid(grid_len = None)
-        hist, _, _ = np.histogram2d(self.transcripts["global_x"], self.transcripts["global_y"], bins = [x_bins, y_bins])
-        area = np.count_nonzero(hist) * (self.grid_len ** 2)
-        return area
+        if not hasattr(self, "_cached_area"):
+            x_bins, y_bins = self.construct_grid(grid_len = None)
+            hist, _, _ = np.histogram2d(self.transcripts["global_x"], self.transcripts["global_y"], bins = [x_bins, y_bins])
+            self._cached_area = np.count_nonzero(hist) * (self.grid_len ** 2)
+        return self._cached_area
     
     
     # [INNER] calculate optimal min_samples, input for dbscan()
@@ -72,24 +75,26 @@ class mcDETECT:
         return optimal_m
     
     
-    # [INTERMEDIATE] dictionary, low- and high-in-nucleus spheres for each synaptic marker
-    def dbscan(self, target_names = None, write_csv = False, write_path = "./"):
+    # [INTERMEDIATE] dictionary, low- and high-in-nucleus spheres for each granule marker
+    def dbscan(self, target_names = None, record_cell_id = False, write_csv = False, write_path = "./"):
         
         if self.type != "Xenium":
             z_grid = list(self.transcripts["global_z"].unique())
             z_grid.sort()
         
         if target_names is None:
-            target_names = self.syn_genes
+            target_names = self.gnl_genes
+        
         transcripts = self.transcripts[self.transcripts["target"].isin(target_names)]
+        grouped = {g: df for g, df in transcripts.groupby("target")}
         
         num_individual, data_low, data_high = [], {}, {}
         
         for j in target_names:
             
             # split transcripts
-            target = transcripts[transcripts["target"] == j]
-            others = transcripts[transcripts["target"] != j]
+            target = grouped[j]
+            others = pd.concat([grouped[g] for g in target_names if g != j], ignore_index = True)
             tree = make_tree(d1 = np.array(others["global_x"]), d2 = np.array(others["global_y"]), d3 = np.array(others["global_z"]))
             
             # 3D DBSCAN
@@ -103,17 +108,25 @@ class mcDETECT:
             n_clusters = len(set(labels)) - (1 if -1 in labels else 0)
             
             # iterate over all aggregations
-            sphere_x, sphere_y, sphere_z, layer_z, sphere_r, sphere_size, sphere_comp, sphere_score = [], [], [], [], [], [], [], []
+            cell_id, sphere_x, sphere_y, sphere_z, layer_z, sphere_r, sphere_size, sphere_comp, sphere_score = [], [], [], [], [], [], [], [], []
             
             for k in range(n_clusters):
                 
+                # record cell ids
+                if record_cell_id:
+                    temp = target[labels == k]
+                    temp_cell_id_mode = temp["cell_id"].mode()[0]
+                    cell_id.append(temp_cell_id_mode)
+                
                 # find minimum enclosing spheres
-                temp = target[labels == k]
-                temp_in_nucleus = np.sum(temp["overlaps_nucleus"])
-                temp_size = temp.shape[0]
-                temp = temp[["global_x", "global_y", "global_z"]]
-                temp = temp.drop_duplicates()
-                center, r2 = miniball.get_bounding_ball(np.array(temp), epsilon=1e-8)
+                mask = (labels == k)
+                coords = X[mask]
+                if coords.shape[0] == 0:
+                    continue
+                temp_in_nucleus = np.sum(target["overlaps_nucleus"].values[mask])
+                temp_size = coords.shape[0]
+                coords_unique = np.unique(coords, axis=0)
+                center, r2 = miniball.get_bounding_ball(coords_unique, epsilon=1e-8)
                 if self.type != "Xenium":
                     closest_z = closest(z_grid, center[2])
                 else:
@@ -139,11 +152,13 @@ class mcDETECT:
                 sphere_comp.append(total_comp)
                 sphere_score.append(local_score)
             
-            # basic features for all spheres from each synaptic marker
-            sphere = pd.DataFrame(list(zip(sphere_x, sphere_y, sphere_z, layer_z, sphere_r, sphere_size, sphere_comp, sphere_score)),
-                                  columns = ["sphere_x", "sphere_y", "sphere_z", "layer_z", "sphere_r", "size", "comp", "in_nucleus"])
-            sphere["gene"] = [j] * sphere.shape[0]
-            sphere = sphere.astype({"sphere_x": float, "sphere_y": float, "sphere_z": float, "layer_z": int, "sphere_r": float, "size": float, "comp": float, "in_nucleus": int, "gene": str})
+            # basic features for all spheres from each granule marker
+            sphere = pd.DataFrame(list(zip(sphere_x, sphere_y, sphere_z, layer_z, sphere_r, sphere_size, sphere_comp, sphere_score, [j] * len(sphere_x))),
+                                      columns = ["sphere_x", "sphere_y", "sphere_z", "layer_z", "sphere_r", "size", "comp", "in_nucleus", "gene"])
+            sphere = sphere.astype({"sphere_x": float, "sphere_y": float, "sphere_z": float, "layer_z": float, "sphere_r": float, "size": float, "comp": float, "in_nucleus": float, "gene": str})
+            if record_cell_id:
+                sphere["cell_id"] = cell_id
+                sphere = sphere.astype({"cell_id": str})
             
             # split low- and high-in-nucleus spheres
             sphere_low = sphere[(sphere["sphere_r"] < self.size_thr) & (sphere["in_nucleus"] < self.in_nucleus_thr[0])]
@@ -156,14 +171,14 @@ class mcDETECT:
             num_individual.append(sphere_low.shape[0])
             data_low[target_names.index(j)] = sphere_low
             data_high[target_names.index(j)] = sphere_high
-            print("{} out of {} genes processed!".format(target_names.index(j) + 1, len(target_names)))
+            print(f"{target_names.index(j) + 1} / {len(target_names)} genes processed!")
         
         return np.sum(num_individual), data_low, data_high
     
     
     # [INNER] merge points from two overlapped spheres, input for remove_overlaps()
     def find_points(self, sphere_a, sphere_b):
-        transcripts = self.transcripts[self.transcripts["target"].isin(self.syn_genes)]
+        transcripts = self.transcripts[self.transcripts["target"].isin(self.gnl_genes)]
         tree_temp = make_tree(d1 = np.array(transcripts["global_x"]), d2 = np.array(transcripts["global_y"]), d3 = np.array(transcripts["global_z"]))
         idx_a = tree_temp.query_ball_point([sphere_a["sphere_x"], sphere_a["sphere_y"], sphere_a["sphere_z"]], sphere_a["sphere_r"])
         points_a = transcripts.iloc[idx_a]
@@ -184,7 +199,7 @@ class mcDETECT:
         # find possible overlaps on 2D by r-tree
         idx_b = make_rtree(set_b)
         for i, sphere_a in set_a.iterrows():
-            center_a_3D = (sphere_a.sphere_x, sphere_a.sphere_y, sphere_a.sphere_z)
+            center_a_3D = np.array([sphere_a.sphere_x, sphere_a.sphere_y, sphere_a.sphere_z])
             bounds_a = (sphere_a.sphere_x - sphere_a.sphere_r,
                         sphere_a.sphere_y - sphere_a.sphere_r,
                         sphere_a.sphere_x + sphere_a.sphere_r,
@@ -195,8 +210,8 @@ class mcDETECT:
             for j in possible_overlaps:
                 if j in set_b.index:
                     sphere_b = set_b.loc[j]
-                    center_b_3D = (sphere_b.sphere_x, sphere_b.sphere_y, sphere_b.sphere_z)
-                    dist = math.dist(center_a_3D, center_b_3D)
+                    center_b_3D = np.array([sphere_b.sphere_x, sphere_b.sphere_y, sphere_b.sphere_z])
+                    dist = np.linalg.norm(center_a_3D - center_b_3D)
                     radius_sum = sphere_a.sphere_r + sphere_b.sphere_r
                     radius_diff = sphere_a.sphere_r - sphere_b.sphere_r
 
@@ -227,10 +242,10 @@ class mcDETECT:
         return set_a, set_b
     
     
-    # [INNER] merge spheres from different synaptic markers, input for detect()
+    # [INNER] merge spheres from different granule markers, input for detect()
     def merge_sphere(self, sphere_dict):
         sphere = sphere_dict[0].copy()
-        for j in range(1, len(self.syn_genes)):
+        for j in range(1, len(self.gnl_genes)):
             target_sphere = sphere_dict[j]
             sphere, target_sphere_new = self.remove_overlaps(sphere, target_sphere)
             sphere = pd.concat([sphere, target_sphere_new])
@@ -268,23 +283,19 @@ class mcDETECT:
         # negative control filtering
         nc_transcripts_final = self.transcripts[self.transcripts["target"].isin(nc_genes_final)]
         tree = make_tree(d1 = np.array(nc_transcripts_final["global_x"]), d2 = np.array(nc_transcripts_final["global_y"]), d3 = np.array(nc_transcripts_final["global_z"]))
-        pass_idx = [0] * sphere_low.shape[0]
-        for i in range(sphere_low.shape[0]):
-            temp = sphere_low.iloc[i]
-            nc_idx = tree.query_ball_point([temp["sphere_x"], temp["sphere_y"], temp["sphere_z"]], temp["sphere_r"])
-            if len(nc_idx) == 0:
-                pass_idx[i] = 1
-            elif len(nc_idx) / temp["size"] < self.nc_thr:
-                pass_idx[i] = 2
-        sphere = sphere_low[np.array(pass_idx) != 0]
-        sphere = sphere.reset_index(drop = True)
+        centers = sphere_low[["sphere_x", "sphere_y", "sphere_z"]].to_numpy()
+        radii = sphere_low["sphere_r"].to_numpy()
+        sizes = sphere_low["size"].to_numpy()
+        counts = np.array([len(tree.query_ball_point(c, r)) for c, r in zip(centers, radii)])
+        pass_idx = (counts == 0) | (counts / sizes < self.nc_thr)
+        sphere = sphere_low[pass_idx].reset_index(drop = True)
         return sphere
     
     
-    # [MAIN] dataframe, synapse metadata
-    def detect(self):
+    # [MAIN] dataframe, granule metadata
+    def detect(self, record_cell_id = False):
         
-        _, data_low, data_high = self.dbscan()
+        _, data_low, data_high = self.dbscan(record_cell_id = record_cell_id)
         
         print("Merging spheres...")
         sphere_low, sphere_high = self.merge_sphere(data_low), self.merge_sphere(data_high)
@@ -296,32 +307,44 @@ class mcDETECT:
             return self.nc_filter(sphere_low, sphere_high)
     
     
-    # [MAIN] anndata, synapse spatial transcriptome profile
-    def profile(self, synapse, genes = None, print_itr = False):
+    # [MAIN] anndata, granule spatial transcriptome profile
+    def profile(self, granule, genes = None, print_itr = False):
         
         if genes is None:
             genes = list(self.transcripts["target"].unique())
             transcripts = self.transcripts
         else:
             transcripts = self.transcripts[self.transcripts["target"].isin(genes)]
+        
+        gene_to_idx = {g: i for i, g in enumerate(genes)}
+        gene_array = transcripts["target"].to_numpy()
         tree = make_tree(d1 = np.array(transcripts["global_x"]), d2 = np.array(transcripts["global_y"]), d3 = np.array(transcripts["global_z"]))
         
-        # construct gene count matrix
-        X = np.zeros((len(genes), synapse.shape[0]))
-        for i in range(synapse.shape[0]):
-            temp = synapse.iloc[i]
+        n_gnl = granule.shape[0]
+        n_gene = len(genes)
+        data, row_idx, col_idx = [], [], []
+        
+        # iterate over all granules to count nearby transcripts
+        for i in range(n_gnl):
+            temp = granule.iloc[i]
             target_idx = tree.query_ball_point([temp["sphere_x"], temp["sphere_y"], temp["layer_z"]], temp["sphere_r"])
-            target_trans = transcripts.iloc[target_idx]
-            target_gene = list(target_trans["target"])
-            for j in np.unique(target_gene):
-                X[genes.index(j), i] = target_gene.count(j)
-            if (print_itr) & (i % 5000 == 0):
-                print("{} out of {} synapses profiled!".format(i, synapse.shape[0]))
+            if not target_idx:
+                continue
+            local_genes = gene_array[target_idx]    # extract genes for those nearby transcripts
+            counts = Counter(local_genes)           # count how many times each gene occurs
+            for g, cnt in counts.items():           # append nonzero entries to sparse matrix lists
+                j = gene_to_idx[g]                  # get gene column index
+                data.append(cnt)                    # nonzero count
+                row_idx.append(i)                   # row index = granule index
+                col_idx.append(j)                   # column index = gene index
+            if print_itr and (i % 5000 == 0):
+                print(f"{i} out of {n_gnl} granules profiled!")
         
-        # construct spatial transcriptome profile
-        adata = anndata.AnnData(X = np.transpose(X), obs = synapse)
-        adata.obs["synapse_id"] = ["syn_{}".format(i) for i in range(synapse.shape[0])]
-        adata.obs["synapse_id"] = adata.obs["synapse_id"].astype(str)
+        # construct sparse spatial transcriptome profile, (n_granules × n_genes)
+        X = csr_matrix((data, (row_idx, col_idx)), shape = (n_gnl, n_gene), dtype = np.float32)
+        adata = anndata.AnnData(X = X, obs = granule.copy())
+        adata.obs["granule_id"] = [f"gnl_{i}" for i in range(n_gnl)]
+        adata.obs = adata.obs.astype({"granule_id": str})
         adata.obs.rename(columns = {"sphere_x": "global_x", "sphere_y": "global_y", "sphere_z": "global_z"}, inplace = True)
         adata.var["genes"] = genes
         adata.var_names = genes
@@ -359,7 +382,7 @@ class mcDETECT:
             count_gene, _, _ = np.histogram2d(target_gene["global_x"], target_gene["global_y"], bins = [x_bins, y_bins])
             X[k_idx, :] = count_gene.flatten()
             if k_idx % 100 == 0:
-                print("{} out of {} genes profiled!".format(k_idx, len(genes)))
+                print(f"{k_idx} out of {len(genes)} genes profiled!")
         
         # spot id
         spot_id = []

{mcdetect-2.0.3.dist-info → mcdetect-2.0.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mcDETECT
-Version: 2.0.3
+Version: 2.0.5
 Summary: Uncovering the dark transcriptome in polarized neuronal compartments with mcDETECT
 Home-page: https://github.com/chen-yang-yuan/mcDETECT
 Author: Chenyang Yuan

mcdetect-2.0.5.dist-info/RECORD

@@ -0,0 +1,8 @@
+mcDETECT/__init__.py,sha256=GbRiy2Zt7JccZDK0rFa5ge7kE9r1L4bERDgQQ1e8QpQ,92
+mcDETECT/model.py,sha256=9V1uNag4tur-JW5MWIPEVyy9yrADxsFR-HpbgU1lkgk,29397
+mcDETECT/utils.py,sha256=0gvqZV7sGi8qvvdC5x9XScyiTXlSfqbUt1zks4t7Xd4,4545
+mcdetect-2.0.5.dist-info/licenses/LICENSE,sha256=uxq-shEWOGTIGVnQLmpElILmfCkuUhFZRAMnZUiKvtg,1070
+mcdetect-2.0.5.dist-info/METADATA,sha256=QE2OBc5Qu18c1iopwx13GkJTp3PEHxpVhX-vo5KccSw,3016
+mcdetect-2.0.5.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+mcdetect-2.0.5.dist-info/top_level.txt,sha256=WwzBojt5U-T2hZ8llO6XgpM9OFIBkWQQldQKu19O8EY,9
+mcdetect-2.0.5.dist-info/RECORD,,

mcdetect-2.0.3.dist-info/RECORD

@@ -1,8 +0,0 @@
-mcDETECT/__init__.py,sha256=SPCzZZOrSFKUNUYRrFbrBWF0FPN6OUzUpRP4zjlfQr0,92
-mcDETECT/model.py,sha256=zEdHqgwTjDi7HxdLW0aPG2j8uLMPiobNu-BcJraAG8g,28047
-mcDETECT/utils.py,sha256=0gvqZV7sGi8qvvdC5x9XScyiTXlSfqbUt1zks4t7Xd4,4545
-mcdetect-2.0.3.dist-info/licenses/LICENSE,sha256=uxq-shEWOGTIGVnQLmpElILmfCkuUhFZRAMnZUiKvtg,1070
-mcdetect-2.0.3.dist-info/METADATA,sha256=1ny7qrjmE9p1Ybgmw3k4QnVJSKlXVJR4nlBNPxj3RCU,3016
-mcdetect-2.0.3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-mcdetect-2.0.3.dist-info/top_level.txt,sha256=WwzBojt5U-T2hZ8llO6XgpM9OFIBkWQQldQKu19O8EY,9
-mcdetect-2.0.3.dist-info/RECORD,,