app.py

import gradio as gr
import torch
import os
from transformers import AutoTokenizer
from models import AffinityPredictor

# Global variables for model and tokenizers
model = None
molecule_tokenizer = None
protein_tokenizer = None
device = None

def load_model():
    """Load the trained model and tokenizers"""
    global model, molecule_tokenizer, protein_tokenizer, device
    
    # Set device
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    print(f"Using device: {device}")
    
    # Load tokenizers
    molecule_tokenizer = AutoTokenizer.from_pretrained("DeepChem/ChemBERTa-77M-MLM")
    protein_tokenizer = AutoTokenizer.from_pretrained("facebook/esm2_t6_8M_UR50D")
    
    # Initialize model with same configuration as training
    model = AffinityPredictor(
        protein_model_name="facebook/esm2_t6_8M_UR50D",
        molecule_model_name="DeepChem/ChemBERTa-77M-MLM",
        hidden_sizes=[1024, 768, 512, 256, 1],
        inception_out_channels=256,
        dropout=0.05
    )
    
    # Load the trained weights
    model_path = "Davis-Final.pth"
    if os.path.exists(model_path):
        checkpoint = torch.load(model_path, map_location=device)
        # Handle different checkpoint formats
        if 'model_state_dict' in checkpoint:
            model.load_state_dict(checkpoint['model_state_dict'])
        elif 'state_dict' in checkpoint:
            model.load_state_dict(checkpoint['state_dict'])
        else:
            model.load_state_dict(checkpoint)
        print("Model loaded successfully!")
    else:
        print(f"Warning: Model file {model_path} not found. Using randomly initialized weights.")
    
    model.to(device)
    model.eval()
    
    return True

def predict_affinity(smiles, sequence):
    """Predict drug-target affinity using the trained model"""
    global model, molecule_tokenizer, protein_tokenizer, device
    
    # Load model if not already loaded
    if model is None:
        try:
            load_model()
        except Exception as e:
            return f"Error loading model: {str(e)}"
    
    # Validate inputs
    if not smiles or not smiles.strip():
        return "Error: Please enter a valid SMILES string"
    
    if not sequence or not sequence.strip():
        return "Error: Please enter a valid protein sequence"
    
    try:
        model.eval()
        
        # Tokenize inputs
        molecule_encoding = molecule_tokenizer(
            [smiles.strip()],
            padding="max_length",
            truncation=True,
            max_length=128,
            return_tensors="pt"
        )
        
        protein_encoding = protein_tokenizer(
            [sequence.strip()],
            padding="max_length",
            truncation=True,
            max_length=1024,
            return_tensors="pt"
        )
        
        # Create batch dictionary
        batch = {
            "molecule_input_ids": molecule_encoding.input_ids.to(device),
            "molecule_attention_mask": molecule_encoding.attention_mask.to(device),
            "protein_input_ids": protein_encoding.input_ids.to(device),
            "protein_attention_mask": protein_encoding.attention_mask.to(device)
        }
        
        # Make prediction
        with torch.no_grad():
            prediction = model(batch)
            affinity_score = prediction.cpu().item()
        
        return f"Predicted Affinity Score: {affinity_score:.4f}"
        
    except Exception as e:
        return f"Error during prediction: {str(e)}"

# Load model on startup
print("Loading model...")
try:
    load_model()
    print("Model loaded successfully!")
except Exception as e:
    print(f"Warning: Could not load model on startup: {e}")

with gr.Blocks(title="Molecule-Protein Affinity Predictor") as demo:
    gr.Markdown("## Molecule–Protein Affinity Prediction")
    gr.Markdown(
        "Enter a **Molecule SMILES string** and a **Protein amino acid sequence** "
        "then click **Predict** to get the affinity score using the StructureFree-DTA model."
    )
    
    gr.Markdown(
        "### Example inputs:\n"
        "**SMILES:** `CC1=C2C=C(C=CC2=NN1)C3=CC(=CN=C3)OCC(CC4=CC=CC=C4)N`\n"
        "\n**Protein:** `MKKFFDSRREQGGSGLGSGSSGGGGSTSGLGSGYIGRVFGIGRQQVTVDEVLAEGGFAIVFLVRTSNGMKCALKRMFVNNEHDLQVCKREIQIMRDLSGHKNIVGYIDSSINNVSSGDVWEVLILMDFCRGGQVVNLMNQRLQTGFTENEVLQIFCDTCEAVARLHQCKTPIIHRDLKVENILLHDRGHYVLCDFGSATNKFQNPQTEGVNAVEDEIKKYTTLSYRAPEMVNLYSGKIITTKADIWALGCLLYKLCYFTLPFGESQVAICDGNFTIPDNSRYSQDMHCLIRYMLEPDPDKRPDIYQVSYFSFKLLKKECPIPNVQNSPIPAKLPEPVKASEAAAKKTQPKARLTDPIPTTETSIAPRQRPKAGQTQPNPGILPIQPALTPRKRATVQPPPQAAGSSNQPGLLASVPQPKPQAPPSQPLPQTQAKQPQAPPTPQQTPSTQAQGLPAQAQATPQHQQQLFLKQQQQQQQPPPAQQQPAGTFYQQQQAQTQQFQAVHPATQKPAIAQFPVVSQGGSQQQLMQNFYQQQQQQQQQQQQQQLATALHQQQLMTQQAALQQKPTMAAGQQPQPQPAAAPQPAPAQEPAIQAPVRQQPKVQTTPPPAVQGQKVGSLTPPSSPKTQRAGHRRILSDVTHSAVFGVPASKSTQLLQAAAAEASLNKSKSATTTPSGSPRTSQQNVYNPSEGSTWNPFDDDNFSKLTAEELLNKDFAKLGEGKHPEKLGGSAESLIPGFQSTQGDAFATTSFSAGTAEKRKGGQTVDSGLPLLSVSDPFIPLQVPDAPEKLIEGLKSPDTSLLLPDLLPMTDPFGSTSDAVIEKADVAVESLIPGLEPPVPQRLPSQTESVTSNRTDSLTGEDSLLDCSLLSNPTTDLLEEFAPTAISAPVHKAAEDSNLISGFDVPEGSDKVAEDEFDPIPVLITKNPQGGHSRNSSGSSESSLPNLARSLLLVDQLIDL`"
    )

    with gr.Row():
        smiles_input = gr.Textbox(
            label="Molecule SMILES",
            placeholder="e.g. CC(=O)OC1=CC=CC=C1C(=O)O",
            lines=2
        )
        sequence_input = gr.Textbox(
            label="Protein Sequence",
            placeholder="e.g. MVLSPADKTNVKAA...",
            lines=5
        )

    predict_button = gr.Button("Predict", variant="primary")
    output = gr.Textbox(label="Affinity Score", interactive=False)

    predict_button.click(
        fn=predict_affinity,
        inputs=[smiles_input, sequence_input],
        outputs=output
    )

if __name__ == "__main__":
    demo.launch()