File size: 20,232 Bytes
f8a4853
 
 
 
b6d1ec8
f8a4853
b6d1ec8
 
 
f8a4853
b6d1ec8
 
 
 
 
 
 
 
f8a4853
 
 
b6d1ec8
 
 
 
 
f8a4853
b6d1ec8
 
f8a4853
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b6d1ec8
 
 
f8a4853
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b6d1ec8
f8a4853
 
 
b6d1ec8
 
 
 
f8a4853
 
 
 
 
 
 
 
 
 
 
 
 
b6d1ec8
f8a4853
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b6d1ec8
 
 
 
 
 
 
 
 
 
 
 
 
 
f8a4853
 
 
 
 
 
 
 
 
 
 
b6d1ec8
f8a4853
 
 
b6d1ec8
f8a4853
 
 
b6d1ec8
 
f8a4853
b6d1ec8
f8a4853
 
 
 
 
b6d1ec8
 
 
 
 
f8a4853
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b6d1ec8
 
 
 
 
 
 
 
 
 
 
 
 
 
f8a4853
 
 
 
 
 
 
 
 
 
 
 
 
b6d1ec8
 
f8a4853
b6d1ec8
 
 
 
f8a4853
b6d1ec8
f8a4853
 
 
 
 
 
 
 
 
 
 
 
 
 
b6d1ec8
f8a4853
 
 
 
 
 
 
b6d1ec8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8a4853
 
 
 
 
 
b6d1ec8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8a4853
 
 
b6d1ec8
f8a4853
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
#!/usr/bin/env python3
# -*- coding: utf-8 -*-

"""
predict_models_dir.py

Predict for all models in models_dir on a folder of FASTA genomes.
Optionally annotate with ground truth from a TSV and compute the same metrics
as in your original script (overall + Isolate + MAG + AUC).

Inputs:
  --genomes_dir   Folder with FASTA files (.fna/.fa/.fasta)
  --models_dir    Folder with model_*.joblib + feature_columns_*.json
  --outdir        Output folder
  --truth_tsv     OPTIONAL: genomes-all_metadata_with_genetic_code_id_noNA.tsv
                  (must contain Genome, Genome_type, Genetic_code_ID)

Ground truth (if provided):
  ALT = Genetic_code_ID != 11
  STD = Genetic_code_ID == 11

Outputs:
  - <outdir>/<model>__pred.csv                          (per model, per genome)
  - <outdir>/all_models_predictions_long.csv            (long: model x genome)
  - <outdir>/prediction_summary.csv                     (ONLY if truth_tsv is provided)
  - <outdir>/top_models_by_pr_auc.txt                   (ONLY if truth_tsv is provided)

Requires:
  - aragorn in PATH (or pass --aragorn)
"""

import os
import re
import json
import time
import argparse
import subprocess
from pathlib import Path
from collections import Counter

import numpy as np
import pandas as pd
from joblib import load as joblib_load

from sklearn.metrics import (
    confusion_matrix,
    accuracy_score,
    precision_score,
    recall_score,
    f1_score,
    roc_auc_score,
    average_precision_score,
)

from sklearn.base import BaseEstimator, ClassifierMixin, clone


# =========================
# PU class (for joblib load)
# =========================
class PUBaggingClassifier(BaseEstimator, ClassifierMixin):
    def __init__(self, base_estimator, n_bags=15, u_ratio=3.0, random_state=42):
        self.base_estimator = base_estimator
        self.n_bags = int(n_bags)
        self.u_ratio = float(u_ratio)
        self.random_state = int(random_state)
        self.models_ = None
        self.classes_ = np.array([0, 1], dtype=int)

    def fit(self, X, y, sample_weight=None):
        y = np.asarray(y).astype(int)
        pos_idx = np.where(y == 1)[0]
        unl_idx = np.where(y == 0)[0]
        if pos_idx.size == 0:
            raise ValueError("PU training requires at least one positive sample (y==1).")

        rng = np.random.RandomState(self.random_state)
        self.models_ = []

        if unl_idx.size == 0:
            m = clone(self.base_estimator)
            try:
                if sample_weight is not None:
                    m.fit(X, y, sample_weight=np.asarray(sample_weight))
                else:
                    m.fit(X, y)
            except TypeError:
                m.fit(X, y)
            self.models_.append(m)
            return self

        k_u = int(min(unl_idx.size, max(1, round(self.u_ratio * pos_idx.size))))
        for _ in range(self.n_bags):
            u_b = rng.choice(unl_idx, size=k_u, replace=(k_u > unl_idx.size))
            idx_b = np.concatenate([pos_idx, u_b])
            X_b = X.iloc[idx_b] if hasattr(X, "iloc") else X[idx_b]
            y_b = y[idx_b]

            sw_b = None
            if sample_weight is not None:
                sw_b = np.asarray(sample_weight)[idx_b]

            m = clone(self.base_estimator)
            try:
                if sw_b is not None:
                    m.fit(X_b, y_b, sample_weight=sw_b)
                else:
                    m.fit(X_b, y_b)
            except TypeError:
                m.fit(X_b, y_b)

            self.models_.append(m)
        return self

    def predict_proba(self, X):
        if not self.models_:
            raise RuntimeError("PUBaggingClassifier not fitted")
        probs = [m.predict_proba(X) for m in self.models_]
        return np.mean(np.stack(probs, axis=0), axis=0)

    def predict(self, X):
        return (self.predict_proba(X)[:, 1] >= 0.5).astype(int)


# =========================
# Feature extraction
# =========================
CODON_RE = re.compile(r"\(([ACGTUacgtu]{3})\)")

def set_single_thread_env():
    os.environ["OMP_NUM_THREADS"] = "1"
    os.environ["OPENBLAS_NUM_THREADS"] = "1"
    os.environ["MKL_NUM_THREADS"] = "1"
    os.environ["VECLIB_MAXIMUM_THREADS"] = "1"
    os.environ["NUMEXPR_NUM_THREADS"] = "1"

def list_fasta_files(genomes_dir: str):
    exts = (".fna", ".fa", ".fasta")
    paths = []
    for fn in os.listdir(genomes_dir):
        p = os.path.join(genomes_dir, fn)
        if not os.path.isfile(p):
            continue
        if fn.endswith(exts):
            paths.append(p)
    return sorted(paths)

def calc_gc_and_tetra(fasta_path):
    bases = ["A", "C", "G", "T"]
    all_kmers = ["".join([a, b, c, d]) for a in bases for b in bases for c in bases for d in bases]
    tetra_counts = {k: 0 for k in all_kmers}

    A = C = G = T = 0
    tail = ""

    with open(fasta_path, "r") as fh:
        for line in fh:
            if line.startswith(">"):
                continue
            s = line.strip().upper().replace("U", "T")
            s = re.sub(r"[^ACGT]", "N", s)
            if not s:
                continue

            seq = tail + s

            for ch in s:
                if ch == "A": A += 1
                elif ch == "C": C += 1
                elif ch == "G": G += 1
                elif ch == "T": T += 1

            for i in range(len(seq) - 3):
                k = seq[i:i+4]
                if "N" in k:
                    continue
                tetra_counts[k] += 1

            tail = seq[-3:] if len(seq) >= 3 else seq

    total_acgt = A + C + G + T
    gc_percent = (float(G + C) / float(total_acgt) * 100.0) if total_acgt > 0 else 0.0

    windows_total = sum(tetra_counts.values())
    denom = float(windows_total) if windows_total > 0 else 1.0
    tetra_freq = {f"tetra_{k}": float(v) / denom for k, v in tetra_counts.items()}

    features = {
        "gc_percent": float(gc_percent),
        "genome_length": float(total_acgt),
    }
    features.update(tetra_freq)
    return features

def run_aragorn(aragorn_bin, fasta_path, out_txt):
    cmd = [aragorn_bin, "-t", "-l", "-gc1", "-w", "-o", out_txt, fasta_path]
    with open(os.devnull, "w") as devnull:
        subprocess.run(cmd, stdout=devnull, stderr=devnull, check=False)

def parse_anticodons_from_aragorn(aragorn_txt):
    counts = Counter()
    if not os.path.isfile(aragorn_txt):
        return counts
    with open(aragorn_txt, "r") as fh:
        for line in fh:
            for m in CODON_RE.finditer(line):
                cod = m.group(1).upper().replace("U", "T")
                if re.fullmatch(r"[ACGT]{3}", cod):
                    counts[cod] += 1
    return counts

def build_ac_features(anticodon_counts):
    bases = ["A", "C", "G", "T"]
    feats = {}
    for a in bases:
        for b in bases:
            for c in bases:
                cod = f"{a}{b}{c}"
                feats[f"ac_{cod}"] = float(anticodon_counts.get(cod, 0))
    return feats

def build_plr_features(ac_features, needed_plr_cols, eps=0.5):
    plr_feats = {}
    for col in needed_plr_cols:
        core = col[len("plr_"):]
        left, right = core.split("__")
        a = ac_features.get(f"ac_{left}", 0.0)
        b = ac_features.get(f"ac_{right}", 0.0)
        plr_feats[col] = float(np.log((a + eps) / (b + eps)))
    return plr_feats

def build_features_for_genome(fasta_path, aragorn_bin, feature_columns, reuse_aragorn=True):
    acc = os.path.splitext(os.path.basename(fasta_path))[0]

    feat_gc_tetra = calc_gc_and_tetra(fasta_path)

    tmp_aragorn = fasta_path + ".aragorn.txt"
    if reuse_aragorn and os.path.isfile(tmp_aragorn):
        try:
            if os.path.getmtime(tmp_aragorn) < os.path.getmtime(fasta_path):
                run_aragorn(aragorn_bin, fasta_path, tmp_aragorn)
        except Exception:
            run_aragorn(aragorn_bin, fasta_path, tmp_aragorn)
    else:
        run_aragorn(aragorn_bin, fasta_path, tmp_aragorn)

    anticodon_counts = parse_anticodons_from_aragorn(tmp_aragorn)
    ac_feats = build_ac_features(anticodon_counts)

    plr_cols = [c for c in feature_columns if c.startswith("plr_")]
    plr_feats = build_plr_features(ac_feats, plr_cols) if plr_cols else {}

    all_feats = {}
    all_feats.update(ac_feats)
    all_feats.update(plr_feats)
    all_feats.update(feat_gc_tetra)

    row = {col: float(all_feats.get(col, 0.0)) for col in feature_columns}
    return acc, row


# =========================
# Ground truth from TSV (optional)
# =========================
def load_truth_tsv(tsv_path: str) -> pd.DataFrame:
    df = pd.read_csv(tsv_path, sep="\t", dtype=str)

    for col in ["Genome", "Genome_type", "Genetic_code_ID"]:
        if col not in df.columns:
            raise ValueError(f"TSV missing column '{col}'. Columns: {list(df.columns)}")

    df["Genome"] = df["Genome"].astype(str)
    df["Genome_type"] = df["Genome_type"].astype(str)
    df["Genetic_code_ID"] = pd.to_numeric(df["Genetic_code_ID"], errors="coerce").astype("Int64")

    # ALT ground truth: != 11
    df["y_true_alt"] = df["Genetic_code_ID"].apply(lambda x: (pd.notna(x) and int(x) != 11)).astype(int)
    df["true_label"] = df["y_true_alt"].map({0: "STD", 1: "ALT"})

    return df[["Genome", "Genome_type", "Genetic_code_ID", "y_true_alt", "true_label"]]


# =========================
# Metrics (only if truth exists)
# =========================
def safe_confusion(y_true, y_pred):
    cm = confusion_matrix(y_true, y_pred, labels=[0, 1])
    tn, fp, fn, tp = int(cm[0,0]), int(cm[0,1]), int(cm[1,0]), int(cm[1,1])
    return tn, fp, fn, tp

def compute_metrics_block(y_true, y_pred, y_score=None):
    y_true = np.asarray(y_true, dtype=int)
    y_pred = np.asarray(y_pred, dtype=int)

    tn, fp, fn, tp = safe_confusion(y_true, y_pred)
    n = int(len(y_true))
    pos = int(np.sum(y_true == 1))

    out = {
        "n": n,
        "positives": pos,
        "tn": tn, "fp": fp, "fn": fn, "tp": tp,
        "accuracy": float(accuracy_score(y_true, y_pred)) if n else np.nan,
        "precision": float(precision_score(y_true, y_pred, zero_division=0)) if n else np.nan,
        "recall": float(recall_score(y_true, y_pred, zero_division=0)) if n else np.nan,
        "f1": float(f1_score(y_true, y_pred, zero_division=0)) if n else np.nan,
        "specificity": float(tn / (tn + fp)) if (tn + fp) > 0 else np.nan,
        "fn_rate": float(fn / (fn + tp)) if (fn + tp) > 0 else np.nan,   # 1 - recall
        "fp_rate": float(fp / (fp + tn)) if (fp + tn) > 0 else np.nan,
    }

    if y_score is not None:
        y_score = np.asarray(y_score, dtype=float)
        if n > 0 and len(np.unique(y_true)) == 2:
            try:
                out["roc_auc"] = float(roc_auc_score(y_true, y_score))
            except Exception:
                out["roc_auc"] = np.nan
            try:
                out["pr_auc"] = float(average_precision_score(y_true, y_score))
            except Exception:
                out["pr_auc"] = np.nan
        else:
            out["roc_auc"] = np.nan
            out["pr_auc"] = np.nan
    else:
        out["roc_auc"] = np.nan
        out["pr_auc"] = np.nan

    return out


# =========================
# Model discovery
# =========================
def find_models(models_dir: Path):
    return sorted(models_dir.glob("model_*.joblib"))

def pick_feature_cols(models_dir: Path, feature_cols_arg: str | None):
    if feature_cols_arg:
        return Path(feature_cols_arg)
    p = models_dir / "feature_columns_64log_gc_tetra.json"
    if p.exists():
        return p
    candidates = sorted(models_dir.glob("feature_columns_*.json"))
    if not candidates:
        raise FileNotFoundError(f"No feature_columns_*.json found in {models_dir}")
    return candidates[0]


# =========================
# Main
# =========================
def main():
    ap = argparse.ArgumentParser(
        description="Predict for all models in a directory; optionally annotate truth from TSV and compute metrics."
    )
    ap.add_argument("--genomes_dir", required=True, help="Folder with FASTA genomes (.fna/.fa/.fasta).")
    ap.add_argument("--models_dir", required=True, help="Folder with model_*.joblib + feature_columns_*.json.")
    ap.add_argument("--outdir", required=True, help="Output folder for CSV predictions.")
    ap.add_argument("--aragorn", default="aragorn", help="Path to ARAGORN binary.")
    ap.add_argument("--feature_cols", default=None, help="Optional: force a specific feature_columns_*.json.")
    ap.add_argument("--reuse_aragorn", action="store_true", help="Reuse *.aragorn.txt if it exists and is fresh.")
    ap.add_argument(
        "--truth_tsv",
        default=None,
        help="OPTIONAL: genomes-all_metadata_with_genetic_code_id_noNA.tsv (Genome, Genome_type, Genetic_code_ID).",
    )
    args = ap.parse_args()

    set_single_thread_env()

    genomes_dir = Path(args.genomes_dir)
    models_dir = Path(args.models_dir)
    outdir = Path(args.outdir)
    outdir.mkdir(parents=True, exist_ok=True)

    fasta_files = list_fasta_files(str(genomes_dir))
    if not fasta_files:
        raise SystemExit(f"No FASTA files found in {genomes_dir}")

    models = find_models(models_dir)
    if not models:
        raise SystemExit(f"No model_*.joblib found in {models_dir}")

    feat_cols_path = pick_feature_cols(models_dir, args.feature_cols)

    print(f"[INFO] Genomes : {len(fasta_files)} in {genomes_dir}")
    print(f"[INFO] Models  : {len(models)} in {models_dir}")
    print(f"[INFO] FeatCols: {feat_cols_path}")
    print(f"[INFO] Truth   : {args.truth_tsv if args.truth_tsv else '(none)'}")

    # Load feature columns
    with open(feat_cols_path, "r") as fh:
        feature_columns = json.load(fh)

    # Optional truth
    truth = None
    if args.truth_tsv:
        truth = load_truth_tsv(args.truth_tsv)

    # Build features once (shared across all models)
    t_feat0 = time.time()
    rows, accs = [], []
    for i, fasta in enumerate(fasta_files, 1):
        if i % 50 == 0 or i == 1 or i == len(fasta_files):
            print(f"[FEAT] {i}/{len(fasta_files)} {os.path.basename(fasta)}")
        acc, feats = build_features_for_genome(
            fasta_path=fasta,
            aragorn_bin=args.aragorn,
            feature_columns=feature_columns,
            reuse_aragorn=args.reuse_aragorn
        )
        accs.append(acc)
        rows.append(feats)

    X = pd.DataFrame(rows, index=accs)[feature_columns]
    print(f"[FEAT] Built X={X.shape} in {(time.time()-t_feat0):.1f}s")

    # Annotation base table (always exists)
    ann = pd.DataFrame({"Genome": accs})

    if truth is not None:
        ann = ann.merge(truth, how="left", on="Genome")
        n_annot = int(ann["y_true_alt"].notna().sum())
        n_missing = int(len(ann) - n_annot)
        print(f"[TRUTH] Annotated: {n_annot}/{len(ann)}  Missing_in_TSV: {n_missing}")
    else:
        # Ensure columns exist for consistent outputs
        ann["Genome_type"] = pd.NA
        ann["Genetic_code_ID"] = pd.NA
        ann["y_true_alt"] = pd.NA
        ann["true_label"] = pd.NA

    long_rows = []
    summary_rows = []

    for mi, model_path in enumerate(models, 1):
        model_name = model_path.stem
        print("\n" + "="*80)
        print(f"[{mi}/{len(models)}] MODEL: {model_path.name}")
        print("="*80)

        t0 = time.time()
        model = joblib_load(model_path)

        # Probabilities (if possible)
        proba = None
        if hasattr(model, "predict_proba"):
            try:
                proba = model.predict_proba(X)[:, 1]
            except Exception:
                proba = None

        # Pred class
        if hasattr(model, "predict"):
            try:
                yhat = model.predict(X)
            except Exception:
                yhat = (proba >= 0.5).astype(int) if proba is not None else np.zeros(len(X), dtype=int)
        else:
            yhat = (proba >= 0.5).astype(int) if proba is not None else np.zeros(len(X), dtype=int)

        elapsed = time.time() - t0

        df_pred = ann.copy()
        df_pred["model"] = model_name
        df_pred["y_pred_alt"] = np.asarray(yhat).astype(int)
        df_pred["pred_label"] = df_pred["y_pred_alt"].map({0: "STD", 1: "ALT"})
        df_pred["proba_alt"] = np.asarray(proba, dtype=float) if proba is not None else np.nan

        out_csv = outdir / f"{model_name}__pred.csv"
        df_pred.to_csv(out_csv, index=False)
        print(f"[WRITE] {out_csv} rows={len(df_pred)} time={(elapsed/60):.2f} min")

        # Long output
        keep_cols = ["model", "Genome", "Genome_type", "Genetic_code_ID", "y_true_alt", "y_pred_alt", "proba_alt"]
        keep_cols = [c for c in keep_cols if c in df_pred.columns]
        long_rows.extend(df_pred[keep_cols].to_dict(orient="records"))

        # Metrics only if we have truth annotations
        if truth is not None:
            df_eval = df_pred[df_pred["y_true_alt"].notna()].copy()
            if df_eval.shape[0] == 0:
                print("[METRICS] No annotated genomes for this run (truth TSV did not match Genome names).")
                continue

            y_true = df_eval["y_true_alt"].astype(int).values
            y_pred = df_eval["y_pred_alt"].astype(int).values
            y_score = df_eval["proba_alt"].astype(float).values if proba is not None else None

            overall = compute_metrics_block(y_true, y_pred, y_score=y_score)

            def subset_metrics(gen_type: str):
                sub = df_eval[df_eval["Genome_type"] == gen_type]
                if sub.shape[0] == 0:
                    return None
                yt = sub["y_true_alt"].astype(int).values
                yp = sub["y_pred_alt"].astype(int).values
                ys = sub["proba_alt"].astype(float).values if proba is not None else None
                return compute_metrics_block(yt, yp, y_score=ys)

            iso = subset_metrics("Isolate")
            mag = subset_metrics("MAG")

            srow = {
                "model": model_name,
                "model_file": str(model_path),
                "feature_cols": str(feat_cols_path),
                "n_genomes_total": int(len(df_pred)),
                "n_annotated": int(df_eval.shape[0]),
                "n_missing_truth": int(len(df_pred) - df_eval.shape[0]),
                "elapsed_sec": float(elapsed),
                "elapsed_min": float(elapsed/60.0),
            }

            for k, v in overall.items():
                srow[f"overall_{k}"] = v

            if iso is not None:
                for k, v in iso.items():
                    srow[f"isolate_{k}"] = v

            if mag is not None:
                for k, v in mag.items():
                    srow[f"mag_{k}"] = v

            summary_rows.append(srow)

    # Write combined outputs
    long_csv = outdir / "all_models_predictions_long.csv"
    pd.DataFrame(long_rows).to_csv(long_csv, index=False)
    print(f"\n[WRITE] {long_csv} rows={len(long_rows)}")

    if truth is not None:
        summary_csv = outdir / "prediction_summary.csv"
        df_sum = pd.DataFrame(summary_rows)
        df_sum.to_csv(summary_csv, index=False)
        print(f"[WRITE] {summary_csv} rows={len(df_sum)}")

        # Top models report
        if not df_sum.empty and "overall_pr_auc" in df_sum.columns:
            df_rank = df_sum.sort_values(["overall_pr_auc", "overall_roc_auc"], ascending=False, na_position="last")
            report_path = outdir / "top_models_by_pr_auc.txt"
            cols = [
                "model",
                "n_annotated",
                "overall_positives",
                "overall_precision",
                "overall_recall",
                "overall_f1",
                "overall_pr_auc",
                "overall_roc_auc",
                "isolate_fn", "isolate_fp", "mag_fn", "mag_fp",
                "elapsed_min",
            ]
            cols = [c for c in cols if c in df_rank.columns]
            with open(report_path, "w", encoding="utf-8") as f:
                f.write("Top models by overall PR-AUC (ALT = Genetic_code_ID != 11)\n")
                f.write(df_rank[cols].head(25).to_string(index=False))
                f.write("\n")
            print(f"[WRITE] {report_path}")

    print("[DONE]")


if __name__ == "__main__":
    main()