Minor bug fixes

CELLULAR_CL/__init__.py

@@ -15,6 +15,7 @@ def train(adata,
           target_key: str,
           batch_key: str,
           latent_dim: int=100,
+          HVG: bool=True,
           HVGs: int=2000,
           model_path: str="trained_models/",
           train_classifier: bool=False,
@@ -60,7 +61,10 @@ def train(adata,
 
     latent_dim (int, optional)
         Dimension of latent space produced by CELLULAR. Default is 100.
-
+
+    HVG (bool, optional)
+        A boolean for whether to filter for highly variable genes or not
+
     HVGs (int, optional)
         Number of highly variable genes (HVGs) to select as input to CELLULAR. Default is 2000.
 
@@ -171,7 +175,11 @@ def train(adata,
     -------
     None
     """
-
+
+    # If not using HVGs set of HVGs to the number of genes
+    if not HVG:
+        HVGs = adata.n_vars
+
     # Raise error if the number of HVGs is not possible to achieve
     if adata.n_vars < HVGs:
         raise ValueError('Number of genes in adata is less than number of HVGs specified to be used.')
@@ -183,7 +191,7 @@ def train(adata,
     # Initiate training class
     train_env = trainer_fun.train_module(data_path=adata,
                                          save_model_path=model_path,
-                                         HVG=True,
+                                         HVG=HVG,
                                          HVGs=HVGs,
                                          target_key=target_key,
                                          batch_keys=[batch_key],
@@ -684,4 +692,4 @@ def rep_seed(seed: int = 42):
     np.random.seed(seed)
     random.seed(seed)
     torch.backends.cudnn.deterministic = True
-    torch.backends.cudnn.benchmark = False
+    torch.backends.cudnn.benchmark = False

CELLULAR_CL/functions/train.py

@@ -164,15 +164,15 @@ def cell_type_centroid_distances(self, n_components: int=100):
         # Step 2: Calculate centroids for each cell type cluster of each batch effect
         centroids = {}
         for batch_effect in adata.obs[self.batch_keys[0]].unique():
-            for cell_type in adata.obs['cell_type'].unique():
-                mask = (adata.obs[self.batch_keys[0]] == batch_effect) & (adata.obs['cell_type'] == cell_type)
+            for cell_type in adata.obs[self.target_key].unique():
+                mask = (adata.obs[self.batch_keys[0]] == batch_effect) & (adata.obs[self.target_key] == cell_type)
                 centroid = np.mean(adata_pca[mask], axis=0)
                 centroids[(batch_effect, cell_type)] = centroid
 
         # Step 3: Calculate the average centroid distance between all batch effects
-        average_distance_matrix = np.zeros((len(adata.obs['cell_type'].unique()), len(adata.obs['cell_type'].unique())))
-        for i, cell_type_i in enumerate(adata.obs['cell_type'].unique()):
-            for j, cell_type_j in enumerate(adata.obs['cell_type'].unique()):
+        average_distance_matrix = np.zeros((len(adata.obs[self.target_key].unique()), len(adata.obs[self.target_key].unique())))
+        for i, cell_type_i in enumerate(adata.obs[self.target_key].unique()):
+            for j, cell_type_j in enumerate(adata.obs[self.target_key].unique()):
                 distances = []
                 for batch_effect in adata.obs[self.batch_keys[0]].unique():
                     centroid_i = torch.tensor(centroids[(batch_effect, cell_type_i)], dtype=torch.float32, requires_grad=False)
@@ -188,7 +188,7 @@ def cell_type_centroid_distances(self, n_components: int=100):
                 average_distance_matrix[i, j] = average_distance
 
         # Convert average_distance_matrix into a DataFrame
-        average_distance_df = pd.DataFrame(average_distance_matrix, index=self.label_encoder.transform(adata.obs['cell_type'].unique()), columns=self.label_encoder.transform(adata.obs['cell_type'].unique()))
+        average_distance_df = pd.DataFrame(average_distance_matrix, index=self.label_encoder.transform(adata.obs[self.target_key].unique()), columns=self.label_encoder.transform(adata.obs[self.target_key].unique()))
 
         # Replace NaN values with 0
         average_distance_df = average_distance_df.fillna(0)
@@ -1021,8 +1021,9 @@ def train_model(self,
 
                 # Update learning rate
                 lr_scheduler.step()
-        except:
+        except Exception as e:
             print(f"**Training forced to finish early due to error during training**")
+            print(e)
 
         print()
         print(f"**Finished training**")
@@ -1151,7 +1152,8 @@ def rep_seed(seed):
                                     batch_keys=self.batch_keys, 
                                     temperature=init_temperature, 
                                     min_temperature=min_temperature, 
-                                    max_temperature=max_temperature)
+                                    max_temperature=max_temperature,
+                                    device = device)
 
         # Define Adam optimer
         optimizer = optim.Adam([{'params': model_step_1.parameters(), 'lr': init_lr}, {'params': loss_module.parameters(), 'lr': init_lr}], weight_decay=5e-5)

requirements.txt

@@ -25,7 +25,7 @@ joblib==1.3.2
 jupyter_client==8.6.0
 jupyter_core==5.7.1
 kiwisolver==1.4.5
-llvmlite==0.42.0
+llvmlite==0.43.0
 Mako==1.3.2
 MarkupSafe==2.1.5
 matplotlib==3.8.3
@@ -34,7 +34,7 @@ mpmath==1.3.0
 natsort==8.4.0
 nest-asyncio==1.6.0
 networkx==3.2.1
-numba==0.59.0
+numba==0.60.0
 numpy==1.26.4
 optuna==3.5.0
 packaging==23.2
@@ -53,7 +53,7 @@ python-dateutil==2.9.0.post0
 pytz==2024.1
 PyYAML==6.0.1
 pyzmq==25.1.2
-scanpy==1.9.8
+scanpy==1.10.0
 scikit-learn==1.4.1.post1
 scipy==1.12.0
 seaborn==0.13.2