scripts/geometric_pose_classification.py

"""Classify each Chroma pose as surface-binding or buried inside the target.

Runs the surface-pose filter from ``utils.pose_filtering`` over the pose PDBs
on disk and writes one verdict per (target, length, pose) to
``benchmark/geometric_pose_classification.csv``. The verdict depends only on the
backbone, so it applies to every sequence designed on that pose.
"""

from __future__ import annotations

import re
from concurrent.futures import ProcessPoolExecutor, as_completed
from pathlib import Path

import pandas as pd

import sys

ROOT = Path(__file__).resolve().parents[2]
if str(ROOT) not in sys.path:
    sys.path.insert(0, str(ROOT))

from utils.pose_filtering import (  # noqa: E402
    DEFAULT_SURFACE_FILTER,
    passes_surface_pose_filter,
    surface_pose_metrics,
)

BENCH = ROOT / "benchmark"

# (target, peptide_mode, designs_root).  ``designs_root`` is the directory that
# holds ``len_*/Poses/Binder_{D,L}_pose_*.pdb``.
SCAN_DIRS = [
    ("TNFalpha", "D_peptide", BENCH / "new_targets" / "designs" / "TNFalpha" / "D_peptide"),
    ("CXCR2", "D_peptide", BENCH / "new_targets" / "designs" / "CXCR2" / "D_peptide"),
    ("CXCR4", "D_peptide", BENCH / "new_targets" / "designs" / "CXCR4" / "D_peptide"),
    ("PDL1", "D_peptide", BENCH / "PDL1" / "D_peptide"),
    ("PDL1", "L_peptide", BENCH / "PDL1" / "L_peptide"),
    ("MDM2", "D_peptide", BENCH / "MDM2" / "D_peptide"),
    ("MDM2", "L_peptide", BENCH / "MDM2" / "L_peptide"),
    ("IL23R", "D_peptide", BENCH / "IL23R" / "D_peptide"),
    ("IL23R", "L_peptide", BENCH / "IL23R" / "L_peptide"),
]

POSE_RE = re.compile(r"Binder_[DL]_pose_(\d+)\.pdb$", re.IGNORECASE)


def discover() -> list[dict]:
    rows = []
    for target, mode, root in SCAN_DIRS:
        if not root.exists():
            print(f"  [skip] missing: {root}")
            continue
        for pose_pdb in sorted(root.glob("len_*/Poses/Binder_[DL]_pose_*.pdb")):
            m = POSE_RE.search(pose_pdb.name)
            if not m:
                continue
            length = int(pose_pdb.parents[1].name.split("_")[1])
            pose = int(m.group(1))
            rows.append(
                {
                    "target": target,
                    "peptide_mode": mode,
                    "length": length,
                    "pose": pose,
                    "pose_pdb": str(pose_pdb.relative_to(ROOT)),
                }
            )
    return rows


def classify_one(row: dict) -> dict:
    pdb = ROOT / row["pose_pdb"]
    try:
        metrics = surface_pose_metrics(
            pdb,
            binder_chain="B",
            nearby_distance=DEFAULT_SURFACE_FILTER["nearby_distance"],
            contact_distance=DEFAULT_SURFACE_FILTER["contact_distance"],
        )
        passed = passes_surface_pose_filter(
            metrics,
            radial_min=DEFAULT_SURFACE_FILTER["radial_min"],
            outward_nearby_max=DEFAULT_SURFACE_FILTER["outward_nearby_max"],
            occupied_octants_max=DEFAULT_SURFACE_FILTER["occupied_octants_max"],
            min_binder_atoms_within_contact=DEFAULT_SURFACE_FILTER[
                "min_binder_atoms_within_contact"
            ],
        )
        out = dict(row)
        out.update(metrics)
        out["passes_surface"] = bool(passed)
        return out
    except Exception as exc:  # pragma: no cover - defensive
        out = dict(row)
        out["error"] = f"{type(exc).__name__}: {exc}"
        out["passes_surface"] = False
        return out


def main():
    jobs = discover()
    print(f"Discovered {len(jobs)} pose PDBs to classify")
    results = []
    with ProcessPoolExecutor(max_workers=16) as ex:
        futs = {ex.submit(classify_one, j): j for j in jobs}
        done = 0
        for fut in as_completed(futs):
            results.append(fut.result())
            done += 1
            if done % 500 == 0:
                print(f"  ...{done}/{len(jobs)}")

    df = pd.DataFrame(results)
    metric_cols = [
        "radial_percentile",
        "outward_nearby_fraction",
        "occupied_octants",
        "binder_atoms_within_contact",
        "min_receptor_distance",
        "median_receptor_distance",
    ]
    cols = ["target", "peptide_mode", "length", "pose", "passes_surface"] + metric_cols + ["pose_pdb"]
    for c in cols:
        if c not in df.columns:
            df[c] = pd.NA
    df = df[cols]

    out = BENCH / "geometric_pose_classification.csv"
    df.to_csv(out, index=False)
    print(f"\nSaved {out}  ({len(df)} rows)")

    print("\nSurface-pose verdict by target/mode:")
    summary = (
        df.groupby(["target", "peptide_mode"])
        .agg(
            n_poses=("pose", "size"),
            n_surface=("passes_surface", "sum"),
        )
        .reset_index()
    )
    summary["n_buried"] = summary["n_poses"] - summary["n_surface"]
    summary["pct_surface"] = (summary["n_surface"] / summary["n_poses"] * 100).round(1)
    print(summary.to_string(index=False))

    errors = df[df["radial_percentile"].isna()]
    if len(errors):
        print(f"\n[warn] {len(errors)} poses failed to compute metrics")


if __name__ == "__main__":
    main()