app.py

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
import torch
import torch.nn as nn
import numpy as np
import joblib

# 1. Define the Neural Network Architecture
# Using Sequential to match your .pth file structure exactly
class MedCareDDI_Network(nn.Module):
    def __init__(self, input_dim):
        super(MedCareDDI_Network, self).__init__()
        self.network = nn.Sequential(
            nn.Linear(input_dim, 512),
            nn.BatchNorm1d(512),
            nn.ReLU(),
            nn.Dropout(0.3),
            
            nn.Linear(512, 256),
            nn.BatchNorm1d(256),
            nn.ReLU(),
            nn.Dropout(0.3),
            
            nn.Linear(256, 128),
            nn.BatchNorm1d(128),
            nn.ReLU(),
            nn.Dropout(0.3),
            
            nn.Linear(128, 3) # Output: 0=Major, 1=Minor, 2=Moderate
        )

    def forward(self, x):
        return self.network(x)

# 2. Initialize FastAPI
app = FastAPI(title="MedCare DDI API", version="2.8")

# Configure CORS for your React Frontend
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=False,
    allow_methods=["*"],
    allow_headers=["*"],
)

# 3. Load Global Resources
device = torch.device("cpu")
input_dim = 12642  # 6321 * 2

print("Initializing AI Engine...")
try:
    drug_vectors = joblib.load('drug_vectors.pkl')
    model = MedCareDDI_Network(input_dim)
    model.load_state_dict(torch.load('MedCareDDI_DrugBank_Model.pth', map_location=device))
    # Standard global eval mode
    model.eval()
    print("Success: Model and Vectors loaded.")
except Exception as e:
    print(f"FAILED TO START: {str(e)}")

# Clinical Severity Mapping (0=Major, 1=Minor, 2=Moderate)
SEVERITY_MAP = {
    0: "Major",
    1: "Minor",
    2: "Moderate"
}

class DDIRequest(BaseModel):
    drug_a_id: str
    drug_b_id: str

@app.get("/")
def health_check():
    return {"status": "online", "model": "DrugBank_MultiModal_v2.8"}

@app.post("/predict")
@app.post("/predict/")
async def predict_interaction(request: DDIRequest):
    # Standardize input IDs (Strip whitespace and uppercase)
    d1_id = request.drug_a_id.strip().upper()
    d2_id = request.drug_b_id.strip().upper()

    # Safety Check: Does the drug exist in our biological database?
    if d1_id not in drug_vectors or d2_id not in drug_vectors:
        missing = d1_id if d1_id not in drug_vectors else d2_id
        raise HTTPException(status_code=400, detail=f"Drug ID {missing} not in database.")

    try:
        # STEP 1: Forced Symmetry
        # Alphabetical sorting ensures A+B always equals B+A mathematically
        drug_ids = sorted([d1_id, d2_id])
        v1 = drug_vectors[drug_ids[0]]
        v2 = drug_vectors[drug_ids[1]]

        # STEP 2: Vector Preparation
        # Reshape to (1, 12642) to provide a single sample batch
        combined = np.concatenate([v1, v2]).astype(np.float32).reshape(1, -1)
        input_tensor = torch.from_numpy(combined).to(device)

        # STEP 3: CRITICAL FIX for BatchNorm1d Error
        # Ensure the model is in eval mode right before passing the tensor
        model.eval()

        # STEP 4: Inference
        with torch.no_grad():
            output = model(input_tensor)
            # Apply Softmax to get probabilities
            probabilities = torch.nn.functional.softmax(output, dim=1)[0]
            
            predicted_idx = torch.argmax(probabilities).item()
            confidence = probabilities[predicted_idx].item() * 100

        # STEP 5: Format Results
        return {
            "status": "success",
            "drug_a": drug_ids[0],
            "drug_b": drug_ids[1],
            "severity": SEVERITY_MAP[predicted_idx],
            "confidence": f"{confidence:.2f}%",
            "raw_scores": {
                "Major (0)": f"{probabilities[0].item():.4f}",
                "Minor (1)": f"{probabilities[1].item():.4f}",
                "Moderate (2)": f"{probabilities[2].item():.4f}"
            }
        }

    except Exception as e:
        # Log the specific cause of 500 errors to the HF console
        print(f"RUNTIME ERROR during prediction: {str(e)}")
        raise HTTPException(status_code=500, detail=f"Internal AI Error: {str(e)}")

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)