evaluation/baselines.py

#!/usr/bin/env python3
"""PROTAC-Bench RF+Morgan baseline reproduction.

Runs 65-fold LOTO evaluation with RandomForest on Morgan fingerprints (1024-bit, radius 2).
Saves predictions to baseline_predictions.csv and runs evaluate.py.

Expected result: AUROC ~0.666

Usage:
    python baselines.py
"""
import json
import subprocess
import sys
from pathlib import Path

import numpy as np
import pandas as pd
from rdkit import Chem, RDLogger
from rdkit.Chem import AllChem
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import roc_auc_score

RDLogger.DisableLog('rdApp.*')

def _find_data_dir():
    """Locate data directory: supports repo layout (../data/) or flat copy (.)."""
    candidates = [
        Path(__file__).resolve().parent.parent / "data",
        Path(__file__).resolve().parent / "data",
        Path(__file__).resolve().parent,  # flat: CSV/JSON beside script
    ]
    for d in candidates:
        if (d / "protac_bench.csv").exists() and (d / "loto_folds.json").exists():
            return d
    return candidates[0]  # fallback to original

DATA_DIR = _find_data_dir()
SEED = 42
N_SEEDS = 3


def auroc_safe(y_true, y_prob):
    if len(np.unique(y_true)) < 2:
        return 0.5
    return roc_auc_score(y_true, y_prob)


def compute_morgan(smiles_list, nbits=2048, radius=2):
    X = np.zeros((len(smiles_list), nbits), dtype=np.float32)
    for i, smi in enumerate(smiles_list):
        mol = Chem.MolFromSmiles(str(smi))
        if mol:
            fp = AllChem.GetMorganFingerprintAsBitVect(mol, radius, nBits=nbits)
            X[i] = np.array(fp)
    return X


def main():
    # Load data
    df = pd.read_csv(DATA_DIR / "protac_bench.csv")
    with open(DATA_DIR / "loto_folds.json") as f:
        folds = json.load(f)
    print(f"Dataset: {len(df)} entries, {len(folds)} LOTO folds")

    # Compute Morgan fingerprints
    print("Computing Morgan fingerprints (1024-bit, radius 2)...")
    X = compute_morgan(list(df["smiles"].values), nbits=1024, radius=2)
    labels = df["label"].values

    # LOTO evaluation with 3 seeds
    all_probs = np.zeros(len(df), dtype=np.float64)
    all_counts = np.zeros(len(df), dtype=np.int32)
    fold_results = []

    for uid, fold in sorted(folds.items()):
        te_idx = np.array(fold["test_indices"])
        tr_idx = np.array([i for i in range(len(df)) if i not in set(te_idx)])
        y_te = labels[te_idx]
        seed_aurocs = []
        seed_probs_list = []
        for s in range(N_SEEDS):
            rf = RandomForestClassifier(
                n_estimators=200, max_depth=None, min_samples_leaf=3,
                random_state=SEED + s, n_jobs=-1
            )
            rf.fit(X[tr_idx], labels[tr_idx])
            prob = rf.predict_proba(X[te_idx])
            prob = prob[:, 1] if rf.classes_[1] == 1 else 1 - prob[:, 0]
            seed_probs_list.append(prob)
            seed_aurocs.append(auroc_safe(y_te, prob))
        avg_probs = np.mean(seed_probs_list, axis=0)
        all_probs[te_idx] = avg_probs
        all_counts[te_idx] = 1
        mean_auc = float(np.mean(seed_aurocs))
        fold_results.append({"target": uid, "n": fold["n_entries"], "auroc": mean_auc})
        print(f"  {uid}: n={fold['n_entries']:3d}  AUROC={mean_auc:.4f}")

    # For entries not in any LOTO fold, predict with full model
    missing = all_counts == 0
    if missing.any():
        rf = RandomForestClassifier(
            n_estimators=200, max_depth=None, min_samples_leaf=3,
            random_state=SEED, n_jobs=-1
        )
        rf.fit(X[~missing], labels[~missing])
        prob = rf.predict_proba(X[missing])
        prob = prob[:, 1] if rf.classes_[1] == 1 else 1 - prob[:, 0]
        all_probs[missing] = prob

    # Summary
    aurocs = [r["auroc"] for r in fold_results]
    print(f"\nRF+Morgan baseline: mean AUROC = {np.mean(aurocs):.4f} "
          f"± {np.std(aurocs):.4f} (n={len(fold_results)} folds)")

    # Save predictions
    script_dir = Path(__file__).resolve().parent
    out_path = script_dir / "baseline_predictions.csv"
    pred_df = pd.DataFrame({
        "index": range(len(df)),
        "predicted_probability": all_probs,
    })
    pred_df.to_csv(out_path, index=False)
    print(f"Predictions saved to {out_path}")

    # Run evaluate.py
    eval_script = script_dir / "evaluate.py"
    result_path = script_dir / "baseline_results.json"
    print(f"\nRunning evaluation...")
    subprocess.run([
        sys.executable, str(eval_script),
        "--predictions", str(out_path),
        "--output", str(result_path),
    ], check=True)


if __name__ == "__main__":
    main()