app.py

import gradio as gr
import pandas as pd
import numpy as np

def predict_risk(file):
    """
    Process uploaded gene expression data and predict heart failure risk.
    
    Args:
        file: Uploaded CSV or XLSX file
        
    Returns:
        DataFrame with Sample IDs, Age, and Heart Failure Risk predictions
    """
    try:
        # Read the uploaded file
        if file.name.endswith('.csv'):
            df = pd.read_csv(file.name)
        elif file.name.endswith('.xlsx'):
            df = pd.read_excel(file.name)
        else:
            return pd.DataFrame({"Error": ["Unsupported file format. Please upload .csv or .xlsx"]})
        
        # Step A: Extract the first column as Sample_IDs
        # Handle both named and unnamed first columns
        first_col_name = df.columns[0]
        Sample_IDs = df.iloc[:, 0].values
        
        # Step B: Extract all other columns as Model_Features (the floats)
        Model_Features = df.iloc[:, 1:].values
        
        # ---------------------------------------------------------
        # REAL MODEL LOADING LOGIC (Add this part)
        # ---------------------------------------------------------
        import joblib
        import os

        # Load your model (ensure 'my_model.pkl' is in your Space's files)
        # If your model is named differently, change this filename!
        model_path = "my_model.pkl" 
        
        if os.path.exists(model_path):
            model = joblib.load(model_path)
            
            # Run the prediction on the extracted features
            # This assumes your model outputs a list of lists like [[Age, Risk], [Age, Risk]]
            predictions = model.predict(Model_Features)
            
            # Split the results
            # If your model outputs a different shape, you might need to adjust index [:, 0] or [:, 1]
            Age = predictions[:, 0]
            Heart_Failure_Risk = predictions[:, 1]
            
        else:
            # Fallback if model file is missing (prevents crashing during setup)
            return pd.DataFrame({"Error": ["Model file not found. Please upload 'my_model.pkl'."]})
        
        # ---------------------------------------------------------
        
        # Step 4: Combine results into a new DataFrame
        results_df = pd.DataFrame({
            'Sample_ID': Sample_IDs,
            'Age': Age,
            'Heart_Failure_Risk': np.round(Heart_Failure_Risk, 4)
        })
        
        return results_df
        
    except Exception as e:
        # Return error message as DataFrame
        return pd.DataFrame({"Error": [f"An error occurred: {str(e)}"]})


# Create Gradio Interface
with gr.Blocks(title="Bioinformatics AI Agent - Heart Failure Risk Prediction") as demo:
    gr.Markdown(
        """
        # 🧬 Bioinformatics AI Agent
        ## Heart Failure Risk Prediction from Gene Expression Data
        
        Upload your gene expression data file (.csv or .xlsx) to predict heart failure risk.
        
        **Expected Format:**
        - First column: Sample IDs (can be named or unnamed)
        - Remaining columns: Gene expression values (numeric features)
        """
    )
    
    with gr.Row():
        with gr.Column():
            file_input = gr.File(
                label="Upload Gene Expression Data",
                file_types=[".csv", ".xlsx"],
                type="filepath"
            )
            predict_btn = gr.Button("Predict Risk", variant="primary")
        
        with gr.Column():
            output_dataframe = gr.Dataframe(
                label="Prediction Results",
                headers=["Sample_ID", "Age", "Heart_Failure_Risk"],
                datatype=["str", "number", "number"],
                row_count=10
            )
    
    gr.Markdown(
        """
        ### 📊 Output Columns:
        - **Sample_ID**: Identifier from your input file
        - **Age**: Predicted age (20-90 years)
        - **Heart_Failure_Risk**: Risk score (0-1, where 1 is highest risk)
        
        ---
        *Note: Current predictions are placeholder values. Replace the prediction logic in `app.py` with your trained model.*
        """
    )
    
    # Connect the button to the prediction function
    predict_btn.click(
        fn=predict_risk,
        inputs=file_input,
        outputs=output_dataframe
    )
    
    # Also allow prediction on file upload
    file_input.change(
        fn=predict_risk,
        inputs=file_input,
        outputs=output_dataframe
    )

# Launch the app
if __name__ == "__main__":
    demo.launch()