update more test code for inference

README.md

@@ -82,13 +82,13 @@ pip install -r requirements.txt
 ```
 ### Full Setup
 - Additional access to trained SVM and ElastNet models requires installation of `RAPIDS cuML`, with instructions available from their official [github page](https://github.com/rapidsai/cuml) (**CUDA-capable GPU required**).
-- Optional: pre-compiled Singularity/Apptainer environment (7.52G) is available at [Google drive](https://drive.google.com/file/d/1RJQ9HK0_gsPOhRo5H5ZmH_MYcpJqQD7e/view?usp=sharing) with everything you need (still need CUDA/GPU to load cuML models).
+- Optional: pre-compiled Singularity/Apptainer environment (5.68G) is available at [Google drive](https://drive.google.com/file/d/1ybLJNTC3BITIqBd8IO09nOOm4PKwD4iS/view?usp=sharing) with everything you need (still need CUDA/GPU to load cuML models). The SHA256 for checking is `48619796ef0adc81bc420021821e5ee3d9b2176bf1f564104e06dc1ce56b3498`, check via `shasum -a 256 peptiverse.sif`. 
     ```
     # test
     apptainer exec peptiverse.sif python -c "import sys; print(sys.executable)"
 
     # run inference (see below)
-    apptainer exec peptiverse.sif python inference.py
+    apptainer exec --nv peptiverse.sif python inference.py
     ```
 ## Repository Structure 🌟
 This repo contains important large files for [PeptiVerse](https://huggingface.co/spaces/ChatterjeeLab/PeptiVerse), an interactive app for peptide property prediction. [Paper link.](https://www.biorxiv.org/content/10.64898/2025.12.31.697180v1)
@@ -112,6 +112,14 @@ PeptiVerse/
 ```
 For full data access, please download the corresponding `training_data_cleaned` and `training_classifiers` from [HuggingFace Dataset](https://huggingface.co/datasets/ChatterjeeLab/PeptiVerse_data) and [Zenodo](https://zenodo.org/records/19989009). The current Huggingface repo only hosts best model weights and meta data with splits labels.
 
+```
+1. Download and extract the Zenodo archive.
+2. Download or clone this repository.
+3. Copy/merge the repository `training_classifiers/` contents into the extracted Zenodo `training_classifiers/` directory, preserving the folder structure.
+`rsync -av --ignore-existing training_classifiers/ /path/to/zenodo_extracted/training_classifiers/`
+4. Do not replace the entire Zenodo folder blindly; merge files so that large training outputs from Zenodo and updated best-model weights from this repository coexist.
+```
+
 ## Training Data Collection 🌟
 
 <table>
@@ -241,7 +249,7 @@ python app.py
 Under the `training_data_cleaned`, we provided the generated embeddings in huggingface dataset format. The following scripts are the steps used to generate the data.
 
 ### Dataset integration
-- All processed training datasets are available at [ChatterjeeLab/PeptiVerse_data](https://huggingface.co/datasets/ChatterjeeLab/PeptiVerse_data). 
+- All processed training datasets are available at [ChatterjeeLab/PeptiVerse\_data](https://huggingface.co/datasets/ChatterjeeLab/PeptiVerse_data). 
 - Selective download the data you need with `huggingface-cli`
 ```bash
 huggingface-cli download ChatterjeeLab/PeptiVerse_data \
@@ -681,4 +689,4 @@ If you find this repository helpful for your publications, please consider citin
 	journal = {bioRxiv}
 }
 ```
-To use this repository, you agree to abide by the MIT License.
+To use this repository, you agree to abide by the MIT License.

inference.py

@@ -13,10 +13,6 @@ from tokenizer.my_tokenizers import SMILES_SPE_Tokenizer
 from lightning.pytorch import seed_everything
 seed_everything(1986)
 
-# -----------------------------
-# Manifest
-# -----------------------------
-
 EMB_TAG_TO_FOLDER_SUFFIX = {
     "wt":         "wt",
     "peptideclm": "peptideclm",
@@ -372,17 +368,10 @@ def _mapie_uncertainty(mapie_bundle: dict, score: float,
     if "quantile" in mapie_bundle:
         q = float(mapie_bundle["quantile"])
         return float(score - q), float(score + q)
-        
-    X_dummy = np.zeros((1, 1))
-    result = mapie.predict(X_dummy)
-    if isinstance(result, tuple):
-        intervals = np.asarray(result[1])
-        if intervals.ndim == 3:
-            return float(intervals[0, 0, 0]), float(intervals[0, 1, 0])
-        return float(intervals[0, 0]), float(intervals[0, 1])
+
     raise RuntimeError(
-        f"Cannot extract intervals: unknown MAPIE bundle format. "
-        f"Bundle keys: {list(mapie_bundle.keys())}."
+        f"Unrecognised MAPIE bundle (keys={list(mapie_bundle.keys()) if isinstance(mapie_bundle, dict) else type(mapie_bundle).__name__}). "
+        f"Run repair_mapie_bundles.py to convert."
     )
 
 def affinity_to_class(y: float) -> int:
@@ -870,8 +859,11 @@ class PeptiVersePredictor:
                 mapie_bundle = self.mapie.get((prop_key, col))
                 if mapie_bundle:
                     emb = get_pooled_emb() if mapie_bundle.get("adaptive") else None
-                    lo, hi = _mapie_uncertainty(mapie_bundle, score, emb)
-                    return (lo, hi), "conformal_prediction_interval"
+                    try:
+                        lo, hi = _mapie_uncertainty(mapie_bundle, score, emb)
+                        return (lo, hi), "conformal_prediction_interval"
+                    except RuntimeError as _e:
+                        return None, f"unavailable (legacy bundle: {_e})"
                 # Fall back to seed ensemble std
                 ens = self.ensembles.get((prop_key, col))
                 if ens:
@@ -892,8 +884,11 @@ class PeptiVersePredictor:
             mapie_bundle = self.mapie.get((prop_key, col))
             if mapie_bundle:
                 emb = get_pooled_emb() if mapie_bundle.get("adaptive") else None
-                lo, hi = _mapie_uncertainty(mapie_bundle, score, emb)
-                return (lo, hi), "conformal_prediction_interval"
+                try:
+                    lo, hi = _mapie_uncertainty(mapie_bundle, score, emb)
+                    return (lo, hi), "conformal_prediction_interval"
+                except RuntimeError as _e:
+                    return None, f"unavailable (legacy bundle: {_e})"
             return None, "unavailable (no MAPIE bundle for XGBoost regression)"
 
         # SVR / ElasticNet regression: MAPIE
@@ -901,8 +896,11 @@ class PeptiVersePredictor:
             mapie_bundle = self.mapie.get((prop_key, col))
             if mapie_bundle:
                 emb = get_pooled_emb() if mapie_bundle.get("adaptive") else None
-                lo, hi = _mapie_uncertainty(mapie_bundle, score, emb)
-                return (lo, hi), "conformal_prediction_interval"
+                try:
+                    lo, hi = _mapie_uncertainty(mapie_bundle, score, emb)
+                    return (lo, hi), "conformal_prediction_interval"
+                except RuntimeError as _e:
+                    return None, f"unavailable (legacy bundle: {_e})"
             return None, "unavailable (MAPIE bundle not found)"
 
         # joblib classifiers (SVM, ElasticNet used as classifier)
@@ -1035,9 +1033,13 @@ class PeptiVersePredictor:
                     emb = np.concatenate([t_vec, b_vec], axis=1)
                 else:
                     emb = None
-                lo, hi = _mapie_uncertainty(mapie_bundle, affinity, emb)
-                out["uncertainty"]      = (lo, hi)
-                out["uncertainty_type"] = "conformal_prediction_interval"
+                try:
+                    lo, hi = _mapie_uncertainty(mapie_bundle, affinity, emb)
+                    out["uncertainty"]      = (lo, hi)
+                    out["uncertainty_type"] = "conformal_prediction_interval"
+                except RuntimeError as _e:
+                    out["uncertainty"]      = None
+                    out["uncertainty_type"] = f"unavailable (legacy bundle: {_e})"
             else:
                 out["uncertainty"]      = None
                 out["uncertainty_type"] = "unavailable (no MAPIE bundle found)"
@@ -1045,35 +1047,142 @@ class PeptiVersePredictor:
         return out
 
 if __name__ == "__main__":
-    root = Path(__file__).resolve().parent  # current script folder
+    import traceback, time
 
-    predictor = PeptiVersePredictor(
-        manifest_path=root / "basic_models.txt",
-        classifier_weight_root=root
-    )
-    print(predictor.training_root)
-    print("MAPIE keys:",    list(predictor.mapie.keys()))
-    print("Ensemble keys:", list(predictor.ensembles.keys()))
-
-    seq = "GIGAVLKVLTTGLPALISWIKRKRQQ"
-    smiles = "C(C)C[C@@H]1NC(=O)[C@@H]2CCCN2C(=O)[C@@H](CC(C)C)NC(=O)[C@@H](CC(C)C)N(C)C(=O)[C@H](C)NC(=O)[C@H](Cc2ccccc2)NC1=O"
-
-    print(predictor.predict_property("hemolysis",   "wt",     seq))
-    print(predictor.predict_property("hemolysis",   "smiles",     smiles, uncertainty=False))
-    print(predictor.predict_property("nf",    "wt",     seq, uncertainty=False))
-    print(predictor.predict_property("nf",    "smiles",     smiles, uncertainty=False))
-    print(predictor.predict_binding_affinity("wt",  target_seq=seq, binder_str="GIGAVLKVLT"))
-    print(predictor.predict_binding_affinity("wt",  target_seq=seq, binder_str="GIGAVLKVLT", uncertainty=False))
-    seq1 = "GIGAVLKVLTTGLPALISWIKRKRQQ"
-    seq2 = "ACDEFGHIKLMNPQRSTVWY"
-    
-    r1 = predictor.predict_binding_affinity("wt",  target_seq=seq2, binder_str="GIGAVLKVLT", uncertainty=False)
-    r2 = predictor.predict_property("nf", "wt", seq1, uncertainty=False)
-    r3 = predictor.predict_property("nf", "wt", seq2, uncertainty=False)
-    r4 = predictor.predict_binding_affinity("wt",  target_seq=seq2, binder_str=smiles, uncertainty=False)
-    r5 = predictor.predict_property("halflife",  "smiles", smiles)
-    print(r1)
-    print(r2)
-    print(r3)
-    print(r4)
-    print(r5)
+    root = Path(__file__).resolve().parent
+
+    SEQ    = "GIGAVLKVLTTGLPALISWIKRKRQQ"   
+    SEQ2   = "ACDEFGHIKLMNPQRSTVWY"          
+    BINDER = "GIGAVLKVLT"                    
+    SMILES = ("C(C)C[C@@H]1NC(=O)[C@@H]2CCCN2C(=O)[C@@H](CC(C)C)"
+              "NC(=O)[C@@H](CC(C)C)N(C)C(=O)[C@H](C)NC(=O)"
+              "[C@H](Cc2ccccc2)NC1=O")
+
+    COL_INPUT = {"wt": SEQ, "smiles": SMILES}
+    BINDING_BINDER = {"wt": BINDER, "smiles": SMILES}
+
+    MANIFESTS = [
+        ("best_models.txt",  "best"),
+        ("basic_models.txt", "basic"),
+    ]
+
+    all_results: List[Tuple[str, str, str]] = []   # (label, status, detail)
+
+    for manifest_file, label in MANIFESTS:
+        sep = "=" * 72
+        print(f"\n{sep}")
+        print(f"  MANIFEST: {manifest_file}  [{label}]")
+        print(sep)
+
+        predictor = PeptiVersePredictor(
+            manifest_path=root / manifest_file,
+            classifier_weight_root=root,
+        )
+
+        print(f"\n  Models loaded : {len(predictor.models)}")
+        print(f"  MAPIE keys    : {sorted(predictor.mapie.keys())}")
+        print(f"  Ensemble keys : {sorted(predictor.ensembles.keys())}")
+        print()
+
+        # ── Regular properties ─────────────────────────────────────────────────
+        for prop_key, row in predictor.manifest.items():
+            if prop_key == "binding_affinity":
+                continue
+
+            for col in ("wt", "smiles"):
+                if (prop_key, col) not in predictor.models:
+                    continue
+
+                input_str = COL_INPUT[col]
+                has_mapie = (prop_key, col) in predictor.mapie
+
+                for uncertainty in (False, True):
+                    tag = f"{label}/{prop_key}/{col}/unc={uncertainty}"
+                    try:
+                        t0 = time.time()
+                        r  = predictor.predict_property(
+                            prop_key, col, input_str, uncertainty=uncertainty
+                        )
+                        dt = time.time() - t0
+
+                        score_str = f"score={r['score']:.4f}"
+                        if uncertainty:
+                            u = r.get("uncertainty")
+                            u_type = r.get("uncertainty_type", "")
+                            u_str  = (f"  unc={u[0]:.3f}..{u[1]:.3f}" if isinstance(u, tuple)
+                                      else f"  unc={u:.4f}" if u is not None
+                                      else f"  unc=None")
+                            mapie_flag = " [MAPIE]" if has_mapie else ""
+                            print(f"  [PASS] {tag:<60} {dt:5.2f}s  {score_str}{u_str}  "
+                                  f"({u_type}){mapie_flag}")
+                        else:
+                            print(f"  [PASS] {tag:<60} {dt:5.2f}s  {score_str}")
+
+                        all_results.append((tag, "PASS", ""))
+                    except Exception as e:
+                        print(f"  [FAIL] {tag}")
+                        traceback.print_exc()
+                        all_results.append((tag, "FAIL", str(e)))
+
+        # ── Binding affinity ───────────────────────────────────────────────────
+        prop_key = "binding_affinity"
+        for col in ("wt", "smiles"):
+            if (prop_key, col) not in predictor.models:
+                continue
+
+            binder_str = BINDING_BINDER[col]
+            has_mapie  = (prop_key, col) in predictor.mapie
+
+            for uncertainty in (False, True):
+                tag = f"{label}/{prop_key}/{col}/unc={uncertainty}"
+                try:
+                    t0 = time.time()
+                    r  = predictor.predict_binding_affinity(
+                        col, target_seq=SEQ2, binder_str=binder_str, uncertainty=uncertainty
+                    )
+                    dt = time.time() - t0
+
+                    aff_str = f"affinity={r['affinity']:.3f}"
+                    if uncertainty:
+                        u = r.get("uncertainty")
+                        u_type = r.get("uncertainty_type", "")
+                        u_str  = (f"  unc={u[0]:.3f}..{u[1]:.3f}" if isinstance(u, tuple)
+                                  else f"  unc=None")
+                        mapie_flag = " [MAPIE]" if has_mapie else ""
+                        print(f"  [PASS] {tag:<60} {dt:5.2f}s  {aff_str}{u_str}  "
+                              f"({u_type}){mapie_flag}")
+                    else:
+                        print(f"  [PASS] {tag:<60} {dt:5.2f}s  {aff_str}  "
+                              f"cls={r['class_by_threshold']}")
+
+                    all_results.append((tag, "PASS", ""))
+                except Exception as e:
+                    print(f"  [FAIL] {tag}")
+                    traceback.print_exc()
+                    all_results.append((tag, "FAIL", str(e)))
+
+        # ── MAPIE bundle inventory ─────────────────────────────────────────────
+        print(f"\n  --- MAPIE bundle inventory [{label}] ---")
+        if not predictor.mapie:
+            print("  (none)")
+        for (pk, col), bundle in sorted(predictor.mapie.items()):
+            fmt   = "adaptive" if bundle.get("adaptive") else "plain_quantile"
+            q     = bundle.get("quantile", "?")
+            sigma = "sigma_model=" + type(bundle["sigma_model"]).__name__ if "sigma_model" in bundle else ""
+            print(f"  {label}/{pk}/{col:<30} format={fmt}  q={q}  {sigma}")
+
+    # ── Summary ────────────────────────────────────────────────────────────────
+    sep = "=" * 72
+    print(f"\n{sep}")
+    print("  SUMMARY")
+    print(sep)
+    passed = [r for r in all_results if r[1] == "PASS"]
+    failed = [r for r in all_results if r[1] == "FAIL"]
+    print(f"  PASSED : {len(passed)} / {len(all_results)}")
+    if failed:
+        print(f"  FAILED : {len(failed)}")
+        for tag, _, err in failed:
+            print(f"    ✗ {tag}")
+            print(f"      {err}")
+    else:
+        print("  All tests passed.")

requirements.txt

@@ -1,14 +1,20 @@
-gradio>=4.0.0
+# ── Core inference  ────────────────────────────────
+numpy>=2.0.0                  # cupy 14.x (RAPIDS 25.x dep) requires numpy >= 2.0
+torch>=2.1.0                  
+transformers==4.46.0          
+scikit-learn>=1.3.0
+lightning==2.5.5             
+joblib>=1.3.0
+xgboost>=2.0.0
+SmilesPE                     
+mapie>=0.6.0                  # conformal prediction (fit_mapie_adaptive.py)
+tqdm
 pandas>=2.0.0
-numpy>=1.24.0
+
+# ── App  ────────────────────────────────────────────────────
+gradio>=4.0.0
 plotly>=5.14.0
-torch>=2.0.0
-transformers==4.46.0
-scikit-learn>=1.3.0
-biopython>=1.81
-lightning==2.5.5
-rdkit>=2023.3.1
 seaborn
-SmilesPE
-xgboost
+rdkit>=2023.3.1
+biopython>=1.81
 ipython