Update src/causal/causal_rank.py

src/causal/causal_rank.py

@@ -1,29 +1,3 @@
-"""
-causal_rank.py — Doubly-robust causal ranking of ABC transporters.
-
-Method
-------
-For each transporter gene (column *_expr in causal_table.csv):
-  1. Binarize treatment: high (> median) vs low expression.
-  2. Fit a DR-Learner (EconML) with RandomForest outcome/propensity models.
-  3. Compute Average Treatment Effect (ATE) + 95% bootstrap CI.
-  4. Perform overlap trimming on the test set (propensity 0.05–0.95).
-  5. Run a placebo test (permuted treatment) to assess false-positive rate.
-
-Inputs
-------
-causal_table.csv columns:
-  outcome      : continuous growth / fitness phenotype
-  ethanol_pct, ROS, NaCl_mM, H2O2_uM, PDR1_reg, YAP1_reg  : covariates
-  batch        : categorical batch covariate
-  *_expr       : one column per transporter (treatment variables)
-
-Outputs
--------
-results/causal_section3_snapshot.json
-results/causal_effects.csv           (sorted by ATE descending)
-"""
-
 from pathlib import Path
 
 import numpy as np
@@ -38,7 +12,6 @@ from econml.dr import DRLearner
 from ..utils.io import load_cfg, set_seed, save_json
 
 
-# ── Helpers ───────────────────────────────────────────────────────────────────
 
 COV_CONT = ["ethanol_pct", "ROS", "NaCl_mM", "H2O2_uM", "PDR1_reg", "YAP1_reg"]
 COV_CAT  = ["batch"]
@@ -74,7 +47,6 @@ def _bootstrap_ate(dr_model, X_test, n=300, seed=17):
     return (float(np.percentile(boots, 2.5)), float(np.percentile(boots, 97.5)))
 
 
-# ── Main ──────────────────────────────────────────────────────────────────────
 
 def run_causal_ranking(
     csv_in:   str = "data/processed/causal_table.csv",
@@ -92,14 +64,12 @@ def run_causal_ranking(
     df = pd.read_csv(csv_in)
     y = df["outcome"].astype(float).to_numpy()
 
-    # Covariates
     prep = _build_preprocessor()
     X_df = df[COV_CONT + COV_CAT].copy()
     X_df[COV_CONT] = X_df[COV_CONT].astype(float)
     X_df[COV_CAT]  = X_df[COV_CAT].astype(str)
     X_all = prep.fit_transform(X_df).astype(np.float32)
 
-    # Treatment columns
     T_cols = [c for c in df.columns if c.endswith("_expr")]
     print(f"Running causal ranking for {len(T_cols)} transporters …")
 
@@ -116,11 +86,9 @@ def run_causal_ranking(
         T_all = (df[gene_col].astype(float).to_numpy() > np.median(df[gene_col])).astype(int)
         T_tr, T_te = T_all[idx_tr], T_all[idx_te]
 
-        # ── Fit DR-Learner ────────────────────────────────────────────────────
         dr = _make_dr_learner(seed)
         dr.fit(y_tr, T_tr, X=X_tr)
 
-        # ── Overlap trimming ──────────────────────────────────────────────────
         prop = LogisticRegression(max_iter=1000, class_weight="balanced")
         prop.fit(X_tr, T_tr)
         e_hat = prop.predict_proba(X_te)[:, 1]
@@ -132,7 +100,6 @@ def run_causal_ranking(
         ate_trim = float(dr.ate(X_trimmed))
         ci = _bootstrap_ate(dr, X_trimmed, seed=seed)
 
-        # ── Placebo test (permuted treatment) ─────────────────────────────────
         T_placebo = rng_placebo.permutation(T_tr)
         dr_pl = _make_dr_learner(seed)
         dr_pl.fit(y_tr, T_placebo, X=X_tr)
@@ -147,7 +114,6 @@ def run_causal_ranking(
         }
         print(f"  {gene:15s}  ATE={ate_trim:+.4f}  CI=[{ci[0]:+.4f}, {ci[1]:+.4f}]")
 
-    # ── Save ──────────────────────────────────────────────────────────────────
     eff_df = (
         pd.DataFrame(results).T
         .reset_index()