"""Week 3: run connectivity scoring over all drugs -> ranked_candidates_v1.csv (PLAN §6).

Loads the disease signature + the 300 drug LINCS signatures, computes the weighted-KS
connectivity score per drug, and produces two rankings:
  1. raw connectivity (most negative = strongest reversal = rank 1)
  2. a secondary ranking blending connectivity with a mechanistic prior (sickle-relevant
     target pathways), to temper broad-effect drugs (HDAC/kinase) that dominate raw rankings.

The formal recovery test (ground-truth + negative-control evaluation against the pre-registered
criteria) is Week 4; this script only prints a sanity peek.
"""

from __future__ import annotations

import json
from pathlib import Path

import pandas as pd

import sys
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
from src.scoring import mechanistic_prior, persist_ranking, rank_drugs  # noqa: E402

PROCESSED = Path("data/processed")
PRIOR_LAMBDA = 0.5  # weight of the mechanistic prior in the secondary ranking


def main() -> None:
    sig = json.loads((PROCESSED / "sickle_cell_signature_v1.json").read_text())
    up = [g["gene"] for g in sig["up_regulated"]]
    down = [g["gene"] for g in sig["down_regulated"]]

    sig_matrix = pd.read_parquet(PROCESSED / "lincs_signatures_v1.parquet")  # drug x 978 symbols
    profiles = pd.read_parquet(PROCESSED / "drug_profiles_v1.parquet").set_index("name")

    landmark = set(sig_matrix.columns)
    n_up_ov = len(set(up) & landmark)
    n_down_ov = len(set(down) & landmark)
    print(f"query overlap with 978 landmarks: {n_up_ov} up + {n_down_ov} down = {n_up_ov + n_down_ov}")
    print(f"scoring {len(sig_matrix)} drugs (all scored; 0 without signature)")

    ranked = rank_drugs(up, down, sig_matrix)

    # attach metadata + mechanistic prior
    ranked = ranked.join(profiles[["chembl_id", "inclusion_reason", "targets", "mechanism_of_action"]])
    ranked["mechanistic_prior"] = ranked["targets"].apply(
        lambda t: mechanistic_prior(list(t) if t is not None else [])
    )
    ranked["known_targets"] = ranked["targets"].apply(
        lambda t: "; ".join(t) if t is not None and len(t) else ""
    )
    ranked = ranked.rename(columns={"mechanism_of_action": "mechanism_summary"})

    # secondary, prior-weighted ranking: relevant drugs pushed toward better (more negative)
    ranked["blended_score"] = ranked["normalized_score"] - PRIOR_LAMBDA * ranked["mechanistic_prior"]
    ranked["blended_rank"] = ranked["blended_score"].rank(method="first").astype(int)

    out = ranked.rename_axis("drug_name").reset_index()[[
        "rank", "drug_name", "chembl_id", "connectivity_score", "normalized_score",
        "inclusion_reason", "mechanistic_prior", "blended_rank", "known_targets", "mechanism_summary",
    ]]
    path = persist_ranking(out)
    print(f"wrote {path} ({len(out)} drugs)")

    # --- sanity peek (formal recovery test is Week 4) ---
    print("\n--- sanity peek (raw connectivity rank) ---")
    for gt in ["hydroxyurea", "glutamine"]:
        r = ranked.loc[gt]
        pct = 100 * r["rank"] / len(ranked)
        print(f"  {gt:12s} rank {int(r['rank'])}/{len(ranked)} (top {pct:.0f}%), "
              f"score={r['connectivity_score']:.3f}")
    neg = ranked[ranked["inclusion_reason"] == "negative_control"]
    print(f"  negative controls in bottom half: "
          f"{(neg['rank'] > len(ranked) / 2).sum()}/{len(neg)}")
    print("\n  top 5 raw candidates:")
    for name, r in ranked.nsmallest(5, "connectivity_score").iterrows():
        print(f"    {int(r['rank']):3d}  {name:18s} {r['connectivity_score']:+.3f}  "
              f"[{r['inclusion_reason']}] {r['known_targets'][:50]}")


if __name__ == "__main__":
    main()