herg-xgboost / predict_herg.py
Delower's picture
Upload predict_herg.py with huggingface_hub
568a378 verified
Raw History Blame Contribute Delete
4.36 kB
#!/usr/bin/env python3
"""Standalone inference for the hERG Phase-5 XGBoost models (Delower/herg-xgboost).
Takes SMILES, builds the exact Phase-5 feature blocks with feature_extraction.py,
applies the fitted preprocessing stored inside each model pickle (fold-fitted
ContPrep states: column keep-mask, median imputation, standardisation; log1p for
count blocks), and predicts with the XGBoost model stored in the same pickle.
Usage:
python predict_herg.py --model models/<cls_scaffold.pkl> "CCOc1ccc..." [...]
python predict_herg.py --model models/<reg_scaffold.pkl> --csv compounds.csv
(CSV must have a 'smiles' column; output adds pred_pIC50 / prob_blocker)
Classification pickles output blocker probability at the locked 10 uM cutoff
(call = prob >= 0.5). Regression pickles output predicted pIC50
(derived 10 uM call = pIC50 >= 5.0).
Dependencies: see requirements.txt (rdkit, xgboost, scikit-learn, numpy).
"""
import argparse, csv, json, pickle, sys
import numpy as np
from feature_extraction import featurize_smiles
BLOCK_KIND = {"morgan": "bits", "maccs": "bits", "desc": "cont",
"physchem_extra": "cont", "morgan_counts": "counts",
"ecfp6": "bits", "fcfp4": "bits", "herg": "cont",
"pharm": "counts", "cdk": "bits",
"morgan1024": "bits", "morgan1024_counts": "counts"}
def apply_prep(raw, state):
"""Replicates p5_common.ContPrep.transform from its stored state dict."""
X = np.asarray(raw, dtype=float).reshape(1, -1)
X = np.where(np.isfinite(X), X, np.nan)
if state["log1p"]:
X = np.log1p(np.clip(X, 0, None))
X = X[:, np.asarray(state["keep"], dtype=bool)]
med = np.asarray(state["med"], dtype=float)
X = np.where(np.isnan(X), med, X)
mu = np.asarray(state["mu"], dtype=float)
sd = np.asarray(state["sd"], dtype=float)
return ((X - mu) / sd).astype(np.float32)
def design_row(blocks, combo, preps):
parts = []
for b in combo:
raw = np.asarray(blocks[b], dtype=np.float32)
if b in preps:
parts.append(apply_prep(raw, preps[b]))
else:
parts.append(raw.reshape(1, -1).astype(np.float32))
return np.hstack(parts)
def load_bundle(path):
with open(path, "rb") as fh:
return pickle.load(fh)
def predict(bundle, smiles_list):
out = []
for smi in smiles_list:
_, blocks = featurize_smiles(smi)
if blocks is None:
out.append(None)
continue
X = design_row(blocks, bundle["blocks"], bundle["preps"])
model = bundle["model"]
if bundle["task"] == "cls":
out.append(float(model.predict_proba(X)[0, 1]))
else:
out.append(float(model.predict(X)[0]))
return out
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--model", required=True, help="path to a model pickle from models/")
ap.add_argument("--csv", help="input CSV with a 'smiles' column")
ap.add_argument("--out", help="output CSV (default: stdout summary)")
ap.add_argument("smiles", nargs="*", help="SMILES strings")
args = ap.parse_args()
bundle = load_bundle(args.model)
rows = []
if args.csv:
with open(args.csv) as fh:
for r in csv.DictReader(fh):
rows.append(r)
smiles = [r["smiles"] for r in rows]
else:
smiles = args.smiles
rows = [{"smiles": s} for s in smiles]
if not smiles:
print("no SMILES given", file=sys.stderr)
return 2
preds = predict(bundle, smiles)
col = "prob_blocker" if bundle["task"] == "cls" else "pred_pIC50"
for r, p in zip(rows, preds):
r[col] = "" if p is None else f"{p:.4f}"
if bundle["task"] == "cls":
r["call_10uM"] = "" if p is None else int(p >= 0.5)
else:
r["derived_call_10uM"] = "" if p is None else int(p >= 5.0)
if args.out:
with open(args.out, "w", newline="") as fh:
w = csv.DictWriter(fh, fieldnames=list(rows[0].keys()))
w.writeheader()
w.writerows(rows)
print(f"wrote {args.out} ({len(rows)} rows, model task={bundle['task']}, "
f"protocol={bundle['protocol']})")
else:
for r in rows:
print(json.dumps(r))
return 0
if __name__ == "__main__":
sys.exit(main())