small changes to regressions

scripts/run_grn_evaluation_tw.sh

@@ -2,13 +2,13 @@
 
 # RUN_ID="run_$(date +%Y-%m-%d_%H-%M-%S)"
 
-RUN_ID="scgen_pearson_gb_sub549"
+RUN_ID="pearson_gb"
 resources_dir="s3://openproblems-data/resources/grn"
 publish_dir="s3://openproblems-data/resources/grn/results/${RUN_ID}"
 # grn_models_folder="${resources_dir}/supplementary/grn_models_noised"
 grn_models_folder="${resources_dir}/grn_models"
 reg_type=GB
-subsample=-1
+subsample=-2
 max_workers=20
 
 param_file="./params/${RUN_ID}.yaml"
@@ -24,7 +24,7 @@ grn_names=(
 )
 
 # layers=("pearson" "lognorm" "scgen_pearson" "scgen_lognorm" "seurat_pearson" "seurat_lognorm")
-layers=( "scgen_pearson" )
+layers=( "pearson" )
 
 # Start writing to the YAML file
 cat > $param_file << HERE
@@ -52,17 +52,17 @@ for grn_name in "${grn_names[@]}"; do
   done
 done
 
-# # Append negative control
-# grn_name="negative_control"
-# for layer in "${layers[@]}"; do
-#   append_entry "$grn_name" "" "true"
-# done
+# Append negative control
+grn_name="negative_control"
+for layer in "${layers[@]}"; do
+  append_entry "$grn_name" "" "true"
+done
 
-# # Append positive controls
-# grn_name="positive_control"
-# for layer in "${layers[@]}"; do
-#   append_entry "$grn_name" "true"
-# done
+# Append positive controls
+grn_name="positive_control"
+for layer in "${layers[@]}"; do
+  append_entry "$grn_name" "true"
+done
 
 # Append the remaining output_state and publish_dir to the YAML file
 cat >> $param_file << HERE
@@ -84,7 +84,7 @@ HERE
     --main-script target/nextflow/workflows/run_grn_evaluation/main.nf `
     --workspace 53907369739130 `
     --compute-env 6TeIFgV5OY4pJCk8I0bfOh `
-    --params-file ./params/scgen_pearson_gb_sub549.yaml `
+    --params-file ./params/scgen_pearson_gb_pcs.yaml `
     --config src/common/nextflow_helpers/labels_tw.config
 
 

src/metrics/regression_1/main.py

@@ -191,6 +191,8 @@ def main(par):
         perturbation_data = perturbation_data[mask,:]
     else:
         perturbation_data = perturbation_data[np.random.choice(perturbation_data.n_obs, subsample, replace=False), :]
+
+    print(perturbation_data.shape)
 
     pert_df = pert_df.T  # make it gene*sample
 

src/metrics/regression_1/script_all.py

@@ -52,41 +52,42 @@ def create_negative_control(gene_names) -> np.ndarray:
   for ii, layer in enumerate(layers):
     par["layer"] = layer
     if grn_model=='positive_control':
-      # print('Inferring GRN')
-      # net = create_positive_control(perturbation_data.layers[par["layer"]], groups)
+      print('Inferring positive control')
+      net = create_positive_control(perturbation_data.layers[par["layer"]], groups)
 
-      # net = pd.DataFrame(net, index=gene_names, columns=gene_names)
-      # net = net.loc[:, net.columns.isin(tf_all)]
+      net = pd.DataFrame(net, index=gene_names, columns=gene_names)
+      net = net.loc[:, net.columns.isin(tf_all)]
 
-      # pivoted_net = net.reset_index().melt(id_vars='index', var_name='source', value_name='weight')
+      pivoted_net = net.reset_index().melt(id_vars='index', var_name='source', value_name='weight')
 
-      # pivoted_net = pivoted_net.rename(columns={'index': 'target'})
-      # pivoted_net = pivoted_net[pivoted_net['weight'] != 0]
+      pivoted_net = pivoted_net.rename(columns={'index': 'target'})
+      pivoted_net = pivoted_net[pivoted_net['weight'] != 0]
       par['prediction'] = f"{par['temp_dir']}/{layer}_positive_control.csv"
-      # pivoted_net.to_csv(par['prediction'])
+      print(par['prediction'])
+      pivoted_net.to_csv(par['prediction'])
     elif grn_model=='negative_control':
-      # print('Inferring GRN')
-      # net = create_negative_control(gene_names)
+      print('Inferring negative control')
+      net = create_negative_control(gene_names)
 
-      # pivoted_net = net.reset_index().melt(id_vars='index', var_name='source', value_name='weight')
+      pivoted_net = net.reset_index().melt(id_vars='index', var_name='source', value_name='weight')
 
-      # pivoted_net = pivoted_net.rename(columns={'index': 'target'})
-      # pivoted_net = pivoted_net[pivoted_net['weight'] != 0]
+      pivoted_net = pivoted_net.rename(columns={'index': 'target'})
+      pivoted_net = pivoted_net[pivoted_net['weight'] != 0]
       par['prediction'] = f"{par['temp_dir']}/negative_control.csv"
-      # pivoted_net.to_csv(par['prediction'])
+      pivoted_net.to_csv(par['prediction'])
     else:
       par['prediction'] = f"resources/grn_models/{grn_model}.csv"
-    output = main(par) 
-    output.index = [layer]
+    # output = main(par) 
+    # output.index = [layer]
 
-    if ii == 0:
-      score = output
-    else:
-      score = pd.concat([score, output], axis=0)
+    # if ii == 0:
+    #   score = output
+    # else:
+    #   score = pd.concat([score, output], axis=0)
 
-    print("Write output to file", flush=True)
-    print(grn_model, layer, score)
+    # print("Write output to file", flush=True)
+    # print(grn_model, layer, score)
 
-  print("Write output to file", flush=True)
-  score.to_csv(par['score'])
+  # print("Write output to file", flush=True)
+  # score.to_csv(par['score'])
 

src/metrics/regression_2/main.py

@@ -198,7 +198,7 @@ def static_approach(grn: np.ndarray, n_features: np.ndarray, X: np.ndarray, grou
     # Cross-validate each gene using the inferred GRN to define select input features
     res = cross_validate(reg_type, gene_names, X, groups, grn, n_features, n_jobs=n_jobs)
     mean_r2_scores = np.asarray([res['scores'][j]['avg-r2'] for j in range(len(res['scores']))])
-    mean_r2_scores = mean_r2_scores[mean_r2_scores>-1]
+    mean_r2_scores = mean_r2_scores[mean_r2_scores>-10]
 
     return np.mean(mean_r2_scores)