Add application file

app.py

@@ -0,0 +1,68 @@
+import gradio as gr
+import torch
+import torch.nn as nn
+from transformers import PreTrainedModel, PretrainedConfig
+from tape import ProteinBertForSequenceClassification, TAPETokenizer
+import re
+
+WHITESPACE_RE = re.compile(r"\s+")
+
+def sanitize(s):
+    return WHITESPACE_RE.sub("", str(s).upper())
+
+# Load model directly (avoids trust_remote_code complexity in Spaces)
+# ckpt = torch.load("kinbert_v2_long.pt", map_location="cpu")
+# OR load from your hub repo:
+from huggingface_hub import hf_hub_download
+ckpt_path = hf_hub_download("steveyu323/kinbert_v2_long", "checkpoints/best.pt")
+ckpt = torch.load(ckpt_path, map_location="cpu")
+
+model = ProteinBertForSequenceClassification.from_pretrained("bert-base", num_labels=2)
+model.load_state_dict(ckpt["state_dict"])
+model.eval()
+
+tokenizer = TAPETokenizer(vocab="iupac")
+THRESHOLD = float(ckpt["threshold"])
+MAX_LEN = 1024
+
+def predict(kinase_seq, substrate_seq):
+    kinase_seq = sanitize(kinase_seq)
+    substrate_seq = sanitize(substrate_seq)
+
+    if not kinase_seq or not substrate_seq:
+        return "Please enter both sequences.", None
+
+    kin_toks = tokenizer.tokenize(kinase_seq)
+    sub_toks = tokenizer.tokenize(substrate_seq)
+    toks = kin_toks + ["<sep>"] + sub_toks
+    toks = tokenizer.add_special_tokens(toks)
+    ids = tokenizer.convert_tokens_to_ids(toks)[:MAX_LEN]
+
+    input_ids = torch.tensor([ids], dtype=torch.long)
+    input_mask = torch.ones_like(input_ids)
+
+    with torch.no_grad():
+        (_, _), logits = model(input_ids=input_ids, input_mask=input_mask)
+        prob = float(torch.softmax(logits, dim=-1)[0, 1])
+
+    label = "✅ Interaction" if prob >= THRESHOLD else "❌ No Interaction"
+    return label, round(prob, 4)
+
+demo = gr.Interface(
+    fn=predict,
+    inputs=[
+        gr.Textbox(lines=3, label="Kinase Sequence", placeholder="Paste kinase amino acid sequence..."),
+        gr.Textbox(lines=3, label="Substrate Sequence", placeholder="Paste substrate amino acid sequence..."),
+    ],
+    outputs=[
+        gr.Text(label="Prediction"),
+        gr.Number(label="Interaction Probability"),
+    ],
+    title="KinBERT — Kinase–Substrate Interaction Classifier",
+    description="Predicts whether a kinase will phosphorylate a given substrate sequence.",
+    examples=[
+        ["MGSSHHHHHHSSGENLYFQGH", "ARTKQTARKSTGGKAPRKQL"],
+    ],
+)
+
+demo.launch()

requirements.txt

@@ -0,0 +1,4 @@
+tape-proteins
+transformers
+torch
+gradio