sciv 0.0.89__tar.gz → 0.0.91__tar.gz

{sciv-0.0.89 → sciv-0.0.91}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sciv
-Version: 0.0.89
+Version: 0.0.91
 Summary: Unveiling the pivotal cell types involved in variant function regulation at a single-cell resolution
 Project-URL: github, https://github.com/YuZhengM/sciv
 Author-email: Zheng-Min Yu <yuzmbio@163.com>

{sciv-0.0.89 → sciv-0.0.91}/pyproject.toml

@@ -3,7 +3,7 @@ requires = ["hatchling"]
 build-backend = "hatchling.build"
 [project]
 name = "sciv"
-version = "0.0.89"
+version = "0.0.91"
 authors = [
     { name = "Zheng-Min Yu", email = "yuzmbio@163.com" },
 ]

{sciv-0.0.89 → sciv-0.0.91}/src/sciv/preprocessing/_scvi_.py

@@ -8,6 +8,7 @@ import numpy as np
 import pandas as pd
 
 from anndata import AnnData
+from torch import OutOfMemoryError
 
 from .. import util as ul
 from ..tool import umap, tsne
@@ -126,11 +127,17 @@ def poisson_vi(
     if model_dir is not None:
         if os.path.exists(os.path.join(model_dir, "model.pt")):
             ul.log(__name__).info(f"Due to the existence of file `model.pt`, it is loaded by default.")
+
             try:
                 model = scvi.external.POISSONVI.load(model_dir, adata=adata)
-            except FileExistsError as fee:
-                ul.log(__name__).error(f"File `model.pt` failed to load, you can execute `Poisson VI` again by deleting file `model.pt` ({model_dir}/model.pt).\n {fee}")
-                raise ValueError(f"File `model.pt` failed to load, you can execute `Poisson VI` again by deleting file `model.pt` ({model_dir}/model.pt).")
+            except OutOfMemoryError as ome:
+                ul.log(__name__).warning(f"GPU failed to run, try to switch to CPU running.\n {ome}")
+
+                try:
+                    model = scvi.external.POISSONVI.load(model_dir, adata=adata, accelerator="cpu", devices="cpu")
+                except Exception as e:
+                    ul.log(__name__).error(f"File `model.pt` failed to load, you can execute `Poisson VI` again by deleting file `model.pt` ({model_dir}/model.pt).\n {e}")
+                    raise ValueError(f"File `model.pt` failed to load, you can execute `Poisson VI` again by deleting file `model.pt` ({model_dir}/model.pt).")
         else:
             ul.file_method(__name__).makedirs(model_dir)
             model = __train__()

{sciv-0.0.89 → sciv-0.0.91}/src/sciv/tool/_algorithm_.py

@@ -814,7 +814,9 @@ def _overlap_(regions_sort: DataFrame, variants: DataFrame) -> DataFrame:
 
     variants_overlap_info_list: list = []
 
-    for index, chr_a, start, end in zip(regions_sort["index"], regions_sort["chr"], regions_sort["start"],
+    for index, chr_a, start, end in zip(regions_sort["index"],
+                                        regions_sort["chr"],
+                                        regions_sort["start"],
                                         regions_sort["end"]):
 
         # judge chr
@@ -906,15 +908,15 @@ def overlap_sum(regions: AnnData, variants: dict, trait_info: DataFrame) -> AnnD
     # Peak number
     label_all_size: int = len(label_all)
 
-    # 预先把 peaks 的 index 做成 dict，O(1) 查找
+    # Pre-build a dict of peak indices for O(1) lookup
     label2idx = {lb: i for i, lb in enumerate(label_all)}
 
     trait_names = trait_info["id"].tolist()
     n_trait = len(trait_names)
-    # 提前分配稀疏矩阵，按列填充，最后一次性转成 csc 再转 csr，省内存且快
+    # Pre-allocate sparse matrix, fill column by column, then convert to csc and then csr for efficiency
     row_indices, col_indices, data_vals = [], [], []
 
-    # 检查列存在性一次完成
+    # Check column existence once
     required = {"chr", "start", "end"}
 
     if not required.issubset(regions.var.columns):
@@ -936,7 +938,7 @@ def overlap_sum(regions: AnnData, variants: dict, trait_info: DataFrame) -> AnnD
 
     ul.log(__name__).info("Obtain peak-trait/disease matrix. (overlap variant information)")
 
-    # 外层循环按 trait 并行可再加速，这里先保持单循环
+    # The outer loop can be further accelerated by parallelizing over traits; here we keep it single-threaded for now.
     for col_idx, trait_name in enumerate(tqdm(trait_names)):
         variant: AnnData = variants[trait_name]
         overlap_df: DataFrame = _overlap_(regions_df, variant.obs)
@@ -944,34 +946,26 @@ def overlap_sum(regions: AnnData, variants: dict, trait_info: DataFrame) -> AnnD
         if overlap_df.empty:
             continue
 
-        # 直接拿到 label->variant_id 的列表，省掉 groupby
-        overlap_df = overlap_df.rename(columns={"index": "label"})
-        # 把 label 映射到行号
-        overlap_df = overlap_df[overlap_df["label"].isin(label2idx)]
-
-        if overlap_df.empty:
-            continue
-
-        # 一次性求和：先按 label 分组，把 variant_id 收集成列表
+        # Sum at once: first group by label and collect variant_id into a list
         label_var_ids = (
             overlap_df
-            .groupby("label")["variant_id"]
+            .groupby("index")["variant_id"]
             .apply(list)
             .reset_index()
         )
 
-        # 遍历每个 label，一次性切片求和
+        # Traverse each label, sum once for each variant_id list
         for _, row in label_var_ids.iterrows():
-            label = row["label"]
+            label = row["index"]
             row_idx = label2idx[label]
             var_ids = row["variant_id"]
-            # 切片一次求和，避免逐行切片
+            # Sum once for all variant_ids in the list, avoiding row-by-row slicing
             matrix_sum = variant[var_ids, :].X.sum(axis=0)
 
             if np.isscalar(matrix_sum):
                 matrix_sum = np.asarray(matrix_sum).reshape(1)
 
-            # 收集非零值
+            # Collect non-zero values
             if matrix_sum.size == 1:
                 val = float(matrix_sum)
                 if val != 0:
@@ -985,7 +979,7 @@ def overlap_sum(regions: AnnData, variants: dict, trait_info: DataFrame) -> AnnD
                         col_indices.append(col_idx + t_idx)
                         data_vals.append(float(v))
 
-    # 构建稀疏矩阵，再转 csr
+    # Build sparse matrix, then convert to csr format
     overlap_sparse = sparse.csc_matrix(
         (data_vals, (row_indices, col_indices)),
         shape=(label_all_size, n_trait),

{sciv-0.0.89 → sciv-0.0.91}/src/sciv/tool/_random_walk_.py

@@ -201,7 +201,6 @@ def random_walk(
     if device == 'cpu' or (device == 'auto' and not availability):
         sample_count = seed_cell_weight.shape[1]
 
-        # 使用 joblib.Parallel 并指定 backend='threading' 保证顺序与输入一致
         results = Parallel(n_jobs=n_jobs)(
             delayed(_random_walk_cpu_)(seed_cell_weight[:, i], weight, gamma, epsilon, p)
             for i in tqdm(range(sample_count))
@@ -690,7 +689,7 @@ class RandomWalk:
                 seed_cell_matrix_en[:, i] = seed_cell_en_value / (1 if seed_cell_en_value.sum() == 0 else seed_cell_en_value.sum())
 
         # Parallel processing of all traits and real-time display of progress
-        Parallel(n_jobs=self.n_jobs)(
+        Parallel(n_jobs=self.n_jobs, backend='threading')(
             delayed(_process_single_trait)(i) for i in tqdm(self.trait_range, desc="Obtain progress of seed cells with weights")
         )
 
@@ -955,8 +954,8 @@ class RandomWalk:
             trait_cell_enrichment[:, i][cell_value_credible > self.credible_threshold] = 1
             trait_cell_credible[:, i] = cell_value_credible
 
-        # Process each trait in parallel
-        Parallel(n_jobs=self.n_jobs)(delayed(_process_trait)(i) for i in tqdm(self.trait_range))
+        # Process each trait in parallel, backend='threading' can effectively prevent the read-only parameter issue caused by copying in loky multi-process mode
+        Parallel(n_jobs=self.n_jobs, backend='threading')(delayed(_process_trait)(i) for i in tqdm(self.trait_range))
 
         self.trs_adata.layers[_layer_label_] = to_sparse(trait_cell_enrichment)