Joblib
File size: 5,837 Bytes
4074358
 
 
 
adff1ee
4074358
 
 
 
 
adff1ee
 
4074358
 
 
 
 
adff1ee
 
 
4074358
 
 
 
 
 
 
 
adff1ee
 
 
 
 
 
 
 
 
 
 
 
 
4074358
 
 
adff1ee
 
4074358
adff1ee
 
 
 
 
 
 
4074358
adff1ee
 
4074358
adff1ee
 
 
 
 
4074358
 
 
 
 
 
 
 
adff1ee
 
4074358
 
adff1ee
4074358
adff1ee
 
 
 
 
4074358
 
 
 
 
 
 
 
adff1ee
4074358
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
adff1ee
 
 
 
 
 
 
 
 
 
 
 
 
 
4074358
 
 
 
 
 
adff1ee
 
 
 
 
 
 
 
 
 
4074358
adff1ee
4074358
adff1ee
4074358
 
 
 
 
 
 
 
 
 
 
 
 
adff1ee
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
#!/usr/bin/env python3
from __future__ import annotations

from pathlib import Path
from typing import List, Optional, Tuple

from huggingface_hub import snapshot_download
from inference import (
    PeptiVersePredictor,
    read_best_manifest_csv,
    _parse_model_and_emb,
    EMB_TAG_TO_FOLDER_SUFFIX,
)

# -----------------------------
# Config
# -----------------------------
root = Path(__file__).resolve().parent
MODEL_REPO = "ChatterjeeLab/PeptiVerse"
DEFAULT_ASSETS_DIR = Path(root)
DEFAULT_MANIFEST = Path("./basic_models.txt")

BANNED_MODELS = {"svm", "enet", "svm_gpu", "enet_gpu"}


def _norm_prop_disk(prop_key: str) -> str:
    return "half_life" if prop_key == "halflife" else prop_key


def _resolve_expected_model_dir(
    prop_key: str, model_name: str, emb_tag: Optional[str]
) -> str:
    """
    Resolve the subfolder path inside training_classifiers/<property>/.

    Args:
        prop_key:   manifest property key, e.g. 'hemolysis', 'halflife'.
        model_name: canonical model string, e.g. 'xgb', 'cnn', 'transformer'.
        emb_tag:    embedding tag from manifest: 'wt', 'peptideclm', or 'chemberta'.
                    None means WT (falls back to 'wt').
    """
    disk_prop = _norm_prop_disk(prop_key)
    base = f"training_classifiers/{disk_prop}"

    folder_suffix = EMB_TAG_TO_FOLDER_SUFFIX.get(emb_tag or "wt", emb_tag or "wt")

    if prop_key == "binding_affinity":
        return f"{base}/{model_name}"

    # ------------------------------------------------------------------
    # halflife WT: folders carry a _log suffix
    # ------------------------------------------------------------------
    if prop_key == "halflife" and (emb_tag is None or emb_tag == "wt"):
        if model_name == "transformer":
            return f"{base}/transformer_wt_log"
        if model_name in {"xgb", "xgb_reg"}:
            return f"{base}/xgb_wt_log"

    # ------------------------------------------------------------------
    # Default: <model>_<folder_suffix>
    # e.g. cnn_chemberta, xgb_peptideclm, transformer_wt
    # ------------------------------------------------------------------
    return f"{base}/{model_name}_{folder_suffix}"


def build_allow_patterns_from_manifest(manifest_path: Path) -> List[str]:
    best = read_best_manifest_csv(manifest_path)

    allow: List[str] = []

    for prop_key, row in best.items():
        for col, parsed in [("wt", row.best_wt), ("smiles", row.best_smiles)]:
            if parsed is None:
                continue

            model_name, emb_tag = parsed

            # Replace banned models with xgb
            if model_name in BANNED_MODELS:
                model_name = "xgb"

            model_dir = _resolve_expected_model_dir(prop_key, model_name, emb_tag)

            allow += [
                f"{model_dir}/best_model.json",
                f"{model_dir}/best_model.pt",
                f"{model_dir}/best_model*.joblib",
                f"{model_dir}/best_model*.json",
            ]

    # Deduplicate while preserving order
    seen = set()
    out = []
    for p in allow:
        if p not in seen:
            out.append(p)
            seen.add(p)
    return out


def download_assets(
    repo_id: str,
    manifest_path: Path,
    out_dir: Path,
) -> Path:
    out_dir = out_dir.resolve()
    out_dir.mkdir(parents=True, exist_ok=True)

    allow_patterns = build_allow_patterns_from_manifest(manifest_path)

    snapshot_download(
        repo_id=repo_id,
        local_dir=str(out_dir),
        local_dir_use_symlinks=False,
        allow_patterns=allow_patterns,
    )
    return out_dir


# -----------------------------
# Main
# -----------------------------
def main():
    import argparse

    ap = argparse.ArgumentParser(
        description="Lightweight PeptiVerse inference with on-demand model download."
    )
    ap.add_argument("--repo",     default=MODEL_REPO,              help="HF repo id containing weights/assets.")
    ap.add_argument("--manifest", default=str(DEFAULT_MANIFEST),   help="Path to best_models.txt")
    ap.add_argument("--assets",   default=str(DEFAULT_ASSETS_DIR), help="Where to store downloaded assets")
    ap.add_argument("--device",   default=None,                    help="cuda / cpu / cuda:0, etc")
    ap.add_argument("--dry-run",  action="store_true",             help="Print allow-patterns without downloading")

    ap.add_argument("--property",   default="hemolysis")
    ap.add_argument("--mode",       default="wt", choices=["wt", "smiles"])
    ap.add_argument("--input",      default="GIGAVLKVLTTGLPALISWIKRKRQQ")
    ap.add_argument("--target_seq", default="GIGAVLKVLTTGLPALISWIKRKRQQ")
    ap.add_argument("--binder",     default="GIGAVLKV")
    args = ap.parse_args()

    manifest_path = Path(args.manifest)
    if not manifest_path.exists():
        raise FileNotFoundError(f"Manifest not found: {manifest_path}")

    if args.dry_run:
        patterns = build_allow_patterns_from_manifest(manifest_path)
        print(f"Would download {len(patterns)} patterns:")
        for p in patterns:
            print(" ", p)
        return

    assets_dir = download_assets(
        args.repo, manifest_path=manifest_path, out_dir=Path(args.assets)
    )

    """TEST CODE
    predictor = PeptiVersePredictor(
        manifest_path=manifest_path,
        classifier_weight_root=str(assets_dir),
        device=args.device,
    )
    if args.property == "binding_affinity":
        if not args.target_seq or not args.binder:
            raise ValueError("For binding_affinity, provide --target_seq and --binder.")
        out = predictor.predict_binding_affinity(args.mode, target_seq=args.target_seq, binder_str=args.binder)
    else:
        out = predictor.predict_property(args.property, args.mode, args.input)
    print(out)
    """

if __name__ == "__main__":
    main()