smalle changes to the workflow

janursa · janursa · commit 8e6b9e0661ad · 2025-11-23T20:18:20.000+01:00
diff --git a/scripts/run_all.sh b/scripts/run_all.sh
@@ -1,7 +1,7 @@
 set -e
 
-datasets=( 'replogle' 'op' 'nakatake' 'adamson' 'norman'  'xaira_HEK293T' 'xaira_HCT116'  'parsebioscience' 'ibd_uc' 'ibd_cd' '300BCG' ) #'replogle' 'op' 'nakatake' 'adamson' 'norman'  'xaira_HEK293T' 'xaira_HCT116'  'parsebioscience' 'ibd_uc' 'ibd_cd'  '300BCG') #
-datasets=( 'replogle' 'op'  ) #'replogle' 'op' 'nakatake' 'adamson' 'norman'  'xaira_HEK293T' 'xaira_HCT116'  'parsebioscience' 'ibd_uc' 'ibd_cd' '300BCG') #
+# datasets=( 'replogle' 'op' 'nakatake' 'adamson' 'norman'  'xaira_HEK293T' 'xaira_HCT116'  'parsebioscience' 'ibd_uc' 'ibd_cd' '300BCG' ) #'replogle' 'op' 'nakatake' 'adamson' 'norman'  'xaira_HEK293T' 'xaira_HCT116'  'parsebioscience' 'ibd_uc' 'ibd_cd'  '300BCG') #
+datasets=( 'op' 'replogle'   ) #'replogle' 'op' 'nakatake' 'adamson' 'norman'  'xaira_HEK293T' 'xaira_HCT116'  'parsebioscience' 'ibd_uc' 'ibd_cd' '300BCG') #
 run_local=false # set to true to run locally, false to run on AWS
 
 run_grn_inference=false
@@ -31,29 +31,29 @@ for dataset in "${datasets[@]}"; do
     fi
 
     if [ "$run_grn_evaluation" = true ]; then
-        if [ -f "$trace_file" ]; then
-            dir=$(dirname "$trace_file")
-            base=$(basename "$trace_file" .txt)
-            today=$(date +%Y-%m-%d)
-            cp "$trace_file" "${dir}/${base}_${today}.txt"
-        fi
+        # if [ -f "$trace_file" ]; then
+        #     dir=$(dirname "$trace_file")
+        #     base=$(basename "$trace_file" .txt)
+        #     today=$(date +%Y-%m-%d)
+        #     cp "$trace_file" "${dir}/${base}_${today}.txt"
+        # fi
         
         # if [ "$run_local" = false ]; then
         #     echo "Uploading inference results to AWS"
         #     aws s3 sync  resources/results/$dataset s3://openproblems-data/resources/grn/results/$dataset 
         #     aws s3 sync  s3://openproblems-data/resources/grn/results/$dataset resources/results/$dataset 
         # fi 
-        if [ "$run_local" = false ]; then
-            echo "Downloading inference results from AWS"
-            aws s3 sync  s3://openproblems-data/resources/grn/results/$dataset resources/results/$dataset 
-        fi
-        echo "Running consensus for dataset: $dataset"
-        bash scripts/prior/run_consensus.sh $dataset # run consensus for Regression and ws distance -> needs to be run after adding each method and dataset
+        # if [ "$run_local" = false ]; then
+        #     echo "Downloading inference results from AWS"
+        #     aws s3 sync  s3://openproblems-data/resources/grn/results/$dataset resources/results/$dataset 
+        # fi
+        # echo "Running consensus for dataset: $dataset"
+        # bash scripts/prior/run_consensus.sh $dataset # run consensus for Regression and ws distance -> needs to be run after adding each method and dataset
         
-        if [ "$run_local" = false ]; then
-            echo "Syncing prior results to AWS"
-            aws s3 sync  resources/grn_benchmark/prior s3://openproblems-data/resources/grn/grn_benchmark/prior 
-        fi
+        # if [ "$run_local" = false ]; then
+        #     echo "Syncing prior results to AWS"
+        #     aws s3 sync  resources/grn_benchmark/prior s3://openproblems-data/resources/grn/grn_benchmark/prior 
+        # fi
 
         echo "Running GRN evaluation for dataset: $dataset"
         bash scripts/run_grn_evaluation.sh --dataset=$dataset --run_local=$run_local --build_images=false 
diff --git a/src/methods/geneformer/helper.py b/src/methods/geneformer/helper.py
@@ -928,11 +928,6 @@ def plot_embs(
                 plot_heatmap(embs, emb_dims, label, output_file, kwargs_dict)
 
 
-def parse_args(par):
-    """Parse arguments - placeholder function"""
-    return par
-
-
 def efficient_melting(matrix, gene_names, symmetric=False):
     """Convert matrix to long format"""
     import pandas as pd
diff --git a/src/metrics/all_metrics/helper.py b/src/metrics/all_metrics/helper.py
@@ -10,7 +10,7 @@
 from regression.helper import main as regression
 from ws_distance.helper import main as ws_distance
 from sem.helper import main as sem
-from ar.helper import main as ar
+from anchor_regression.helper import main as ar
 from tf_recovery.helper import main as tf_recovery
 from tf_binding.helper import main as tf_binding
 from rc_tf_act.helper import main as rc_tf_act
@@ -19,7 +19,6 @@
 
 from config import DATASETS_METRICS
 
-# Create a mapping from metric name (string) to function
 METRIC_FUNCTIONS = {
     'regression': regression,
     'ws_distance': ws_distance,
@@ -36,14 +35,14 @@ def main(par):
     dataset_id = ad.read_h5ad(par['evaluation_data'], backed='r').uns['dataset_id']
     rr_store = []
     metrics = DATASETS_METRICS[dataset_id]
+    # metrics = ['gs_recovery', 'tf_binding']
 
     for metric_name in metrics:
         print(f"Computing metric: {metric_name}")
         metric_func = METRIC_FUNCTIONS.get(metric_name)
         if metric_func is None:
             print(f"Warning: No function found for metric '{metric_name}'")
             continue
-       
         
         rr = metric_func(par)
 
diff --git a/src/metrics/anchor_regression/config.vsh.yaml b/src/metrics/anchor_regression/config.vsh.yaml
@@ -1,6 +1,6 @@
 __merge__: ../../api/comp_metric.yaml
 
-name: ar
+name: anchor_regression
 namespace: "metrics"
 info:
   label: Anchor Regression
diff --git a/src/metrics/anchor_regression/helper.py b/src/metrics/anchor_regression/helper.py
diff --git a/src/metrics/anchor_regression/run_global.sh b/src/metrics/anchor_regression/run_global.sh
diff --git a/src/metrics/anchor_regression/run_local.sh b/src/metrics/anchor_regression/run_local.sh
diff --git a/src/metrics/anchor_regression/script.py b/src/metrics/anchor_regression/script.py
diff --git a/src/metrics/gs_recovery/config.vsh.yaml b/src/metrics/gs_recovery/config.vsh.yaml
@@ -91,6 +91,12 @@ arguments:
     required: false
     default: "zero_centered"
     description: "Method for determining pathway activity baseline (zero_centered, permutation, or random_genesets)"
+    
+  - name: "--output_detailed_metrics"
+    type: boolean
+    required: false
+    default: false
+    description: Whether to output detailed per-geneset metrics or only summary metrics
   
 resources:
   - type: python_script
diff --git a/src/metrics/gs_recovery/helper.py b/src/metrics/gs_recovery/helper.py
@@ -792,7 +792,20 @@ def main(par: dict) -> pd.DataFrame:
     """
     
     # Load data
-    print("\n[1/5] Loading data...")
+    pathway_files = {}
+    geneset_mapping = {
+        'geneset_hallmark_2020': 'hallmark_2020',
+        'geneset_kegg_2021': 'kegg_2021',
+        'geneset_reactome_2022': 'reactome_2022',
+        'geneset_go_bp_2023': 'go_bp_2023',
+        'geneset_bioplanet_2019': 'bioplanet_2019',
+        'geneset_wikipathways_2019': 'wikipathways_2019',
+    }
+    
+    for arg_name, geneset_name in geneset_mapping.items():
+        pathway_files[geneset_name] = par[arg_name]
+    
+    par['pathway_files'] = pathway_files
     evaluation_data = ad.read_h5ad(par['evaluation_data'], backed='r')
     all_genes = set(evaluation_data.var_names.tolist())
     prediction = read_prediction(par)
@@ -846,21 +859,25 @@ def main(par: dict) -> pd.DataFrame:
         all_results.append(result_dict)
        
     
-    final_dict = {}
+    detailed_dict = {}
     for result in all_results:
         geneset_name = result['geneset_name']
-        final_dict[f'{geneset_name}_gs_precision'] = result['precision']
-        final_dict[f'{geneset_name}_gs_recall'] = result['recall']
-        final_dict[f'{geneset_name}_gs_f1'] = result['f1']
-        final_dict[f'{geneset_name}_gs_n_active'] = result['n_active_pathways']
+        detailed_dict[f'{geneset_name}_gs_precision'] = result['precision']
+        detailed_dict[f'{geneset_name}_gs_recall'] = result['recall']
+        detailed_dict[f'{geneset_name}_gs_f1'] = result['f1']
+        detailed_dict[f'{geneset_name}_gs_n_active'] = result['n_active_pathways']
     
     # Calculate mean across all gene sets
+    short_dict = {}
     if all_results:
-        final_dict['gs_precision'] = np.mean([r['precision'] for r in all_results])
-        final_dict['gs_recall'] = np.mean([r['recall'] for r in all_results])
-        final_dict['gs_f1'] = np.mean([r['f1'] for r in all_results])
-        final_dict['gs_n_active'] = np.mean([r['n_active_pathways'] for r in all_results])
-    
+        short_dict['gs_precision'] = np.mean([r['precision'] for r in all_results])
+        short_dict['gs_recall'] = np.mean([r['recall'] for r in all_results])
+        short_dict['gs_f1'] = np.mean([r['f1'] for r in all_results])
+        short_dict['gs_n_active'] = np.mean([r['n_active_pathways'] for r in all_results])
+    if par['output_detailed_metrics']:
+        final_dict = {**short_dict, **detailed_dict}
+    else:
+        final_dict = short_dict
     summary_df = pd.DataFrame([final_dict])
     print(summary_df)
     return summary_df
diff --git a/src/metrics/gs_recovery/script.py b/src/metrics/gs_recovery/script.py
@@ -21,7 +21,8 @@
     'max_pathway_size': 500,
     'min_targets': 10,
     'max_targets': 100,  # Top K edges by absolute weight
-    'ulm_baseline_method': 'zero_centered'
+    'ulm_baseline_method': 'zero_centered',
+    'output_detailed_metrics': True
 }
 ## VIASH END
 
@@ -42,21 +43,6 @@
 
 
 if __name__ == "__main__":
-    # Collect geneset files from par dictionary
-    pathway_files = {}
-    geneset_mapping = {
-        'geneset_hallmark_2020': 'hallmark_2020',
-        'geneset_kegg_2021': 'kegg_2021',
-        'geneset_reactome_2022': 'reactome_2022',
-        'geneset_go_bp_2023': 'go_bp_2023',
-        'geneset_bioplanet_2019': 'bioplanet_2019',
-        'geneset_wikipathways_2019': 'wikipathways_2019',
-    }
-    
-    for arg_name, geneset_name in geneset_mapping.items():
-        pathway_files[geneset_name] = par[arg_name]
-    
-    par['pathway_files'] = pathway_files
     
     output = main_helper(par)
     print(output)
diff --git a/src/metrics/tf_binding/config.vsh.yaml b/src/metrics/tf_binding/config.vsh.yaml
@@ -27,6 +27,11 @@ arguments:
     must_exist: false
     required: false
     example: resources_test/grn_benchmark/ground_truth/PBMC.csv 
+  - name: --output_detailed_metrics
+    type: boolean
+    required: false
+    default: false
+    description: Whether to output detailed per-geneset metrics or only summary metrics
   
 
   
diff --git a/src/metrics/tf_binding/helper.py b/src/metrics/tf_binding/helper.py
@@ -141,17 +141,16 @@ def main(par):
         'tfb_recall': tfb_recall_weighted,
         'tfb_f1': tfb_f1
     }
-    
-    # Add GT-specific scores
-    for _, row in result_df.iterrows():
-        gt = row['gt']
-        final_result[f'{gt}_tfb_precision'] = row['tfb_grn_norm']
-        final_result[f'{gt}_tfb_recall'] = row['tfb_all_norm']
-        # Calculate F1 for each GT
-        precision = row['tfb_grn_norm']
-        recall = row['tfb_all_norm']
-        f1_gt = 2 * (precision * recall) / (precision + recall) if (precision + recall) > 0 else 0
-        final_result[f'{gt}_tfb_f1'] = f1_gt
+    if par['output_detailed_metrics']:
+        for _, row in result_df.iterrows():
+            gt = row['gt']
+            final_result[f'{gt}_tfb_precision'] = row['tfb_grn_norm']
+            final_result[f'{gt}_tfb_recall'] = row['tfb_all_norm']
+            # Calculate F1 for each GT
+            precision = row['tfb_grn_norm']
+            recall = row['tfb_all_norm']
+            f1_gt = 2 * (precision * recall) / (precision + recall) if (precision + recall) > 0 else 0
+            final_result[f'{gt}_tfb_f1'] = f1_gt
     
     result_df = pd.DataFrame([final_result])
     
diff --git a/src/metrics/tf_binding/script.py b/src/metrics/tf_binding/script.py
@@ -15,7 +15,8 @@
     'num_workers': 20,
     'tf_all': 'resources/grn_benchmark/prior/tf_all.csv',    
     'score': 'output/score.h5ad',
-    'genes_n': 5000
+    'genes_n': 5000,
+    'output_detailed_metrics': True
 }
 ## VIASH END
 
diff --git a/src/metrics/vc/helper.py b/src/metrics/vc/helper.py
@@ -247,24 +247,15 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
             A = self.A_weights * self.A_mask
         
         if self.inverse:
-            # For inverse transformation, use iterative solve to avoid memory issues
-            # Solve (I - alpha * A.t()) * y = x for y
             ia = self.I - self.alpha * A.t()
-            
-            # Add small regularization to diagonal for numerical stability
             ia = ia + 1e-6 * self.I
-            
-            # Use solve instead of inversion to save memory
             try:
-                # Solve ia * y.t() = x.t() for y.t(), then transpose
                 result = torch.linalg.solve(ia, x.t()).t()
                 return result
             except torch.linalg.LinAlgError:
-                # Fallback: simple linear transformation without inversion
                 print("Warning: Matrix solve failed, using simplified GRN transformation")
                 return torch.mm(x, A)
         else:
-            # Forward transformation: apply GRN directly
             return torch.mm(x, A.t())
 
 
@@ -385,7 +376,6 @@ def evaluate(A, train_data_loader, test_data_loader, n_perturbations: int) -> Tu
         for x, pert, y in train_data_loader:
             x, pert, y = x.to(DEVICE), pert.to(DEVICE), y.to(DEVICE)
             optimizer.zero_grad()
-            # Model now predicts full perturbed expression directly
             y_hat = model(x, pert)
             loss = torch.mean(torch.square(y - y_hat))
             loss.backward()
@@ -410,9 +400,7 @@ def evaluate(A, train_data_loader, test_data_loader, n_perturbations: int) -> Tu
         model.train()
     ss_res = best_ss_res
 
-    # Final evaluation with PDS
     model.eval()
-
     ss_tot = 0
 
     with torch.no_grad():
@@ -427,9 +415,7 @@ def evaluate(A, train_data_loader, test_data_loader, n_perturbations: int) -> Tu
     return best_ss_res, ss_tot
 
 
-
 def main(par):
-    # Load evaluation data
     adata = ad.read_h5ad(par['evaluation_data'])
     assert 'is_control' in adata.obs.columns, "'is_control' column is required in the dataset for perturbation evaluation"
     assert adata.obs['is_control'].sum() > 0, "'is_control' column must contain at least one True value for control samples"
diff --git a/src/utils/config.py b/src/utils/config.py
@@ -108,7 +108,7 @@
         METRICS_DATASETS.setdefault(metric, []).append(dataset)
 
 ORDERED_METRICS = [
-       'r2_precision', 'r2_recall', 'r2_f1',
+       'r_precision', 'r_recall', 'r_f1',
        'ws_precision', 'ws_recall', 'ws_f1',
        'vc', 
        'sem', 
diff --git a/src/utils/util.py b/src/utils/util.py
@@ -99,8 +99,7 @@ def parse_args(par):
     parser.add_argument('--geneset_go_bp_2023', type=str, help='GO BP 2023 geneset GMT file')
     parser.add_argument('--geneset_bioplanet_2019', type=str, help='BioPlanet 2019 geneset GMT file')
     parser.add_argument('--geneset_wikipathways_2019', type=str, help='WikiPathways 2019 geneset GMT file')
-
-   
+    parser.add_argument('--output_detailed_metrics', action='store_true', default=True, help='Output detailed metrics')
     
     args = parser.parse_args()
     for k, v in vars(args).items():
diff --git a/src/workflows/run_grn_evaluation/config.vsh.yaml b/src/workflows/run_grn_evaluation/config.vsh.yaml
@@ -145,7 +145,7 @@ dependencies:
   - name: metrics/vc
   - name: metrics/rc_tf_act
   - name: metrics/sem
-  - name: metrics/ar
+  - name: metrics/anchor_regression
   - name: metrics/tf_binding
   - name: metrics/gs_recovery
   - name: utils/extract_uns_metadata
diff --git a/src/workflows/run_grn_evaluation/main.nf b/src/workflows/run_grn_evaluation/main.nf
@@ -17,7 +17,7 @@ workflow run_wf {
     regression,
     ws_distance,
     tf_recovery,
-    ar,
+    anchor_regression,
     rc_tf_act,
     sem,
     vc,
