app.py

import gradio as gr
import pandas as pd
from datasets import load_dataset, Dataset
from fuzzywuzzy import process
from rdkit import Chem
from rdkit.Chem import AllChem, Draw
import io
import tempfile
import base64
import os
from google import genai
from google.genai import types
from reportlab.pdfgen import canvas
from reportlab.lib.pagesizes import letter
from reportlab.lib.styles import getSampleStyleSheet
from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer, Image
from svglib.svglib import svg2rlg
from huggingface_hub import HfApi

# Load dataset
dataset = load_dataset("smitathkr1/organic_reactions_corrected")
df = dataset['train'].to_pandas()

# Global variable to store the last AI fix for potential database update
last_ai_fix = None

# Precompute unique values for autocomplete
reaction_names = df['corrected_name'].unique().tolist()
all_reactants = []
all_products = []

for _, row in df.iterrows():
    if pd.notna(row['general_reactants']):
        all_reactants.append(row['general_reactants'])
    if pd.notna(row['general_products']):
        all_products.append(row['general_products'])

unique_reactants = list(set(all_reactants))
unique_products = list(set(all_products))

def generate_reaction_svg(name):
    if not name:
        return "Please provide a reaction name."
    # Find the reaction
    result = df[df['corrected_name'].str.lower() == name.lower()]
    if not result.empty:
        row = result.iloc[0]
        reactants_smiles = '.'.join(row['reactants_smiles'])
        products_smiles = '.'.join(row['products_smiles'])
        reaction_smiles = f"{reactants_smiles}>>{products_smiles}"
        try:
            rxn = AllChem.ReactionFromSmarts(reaction_smiles)
            if rxn:
                svg = Draw.ReactionToImage(rxn, useSVG=True)
                return svg
            else:
                return "Failed to parse reaction SMILES."
        except Exception as e:
            return f"Error generating SVG: {str(e)}"
    return "Reaction not found."

def generate_all_reactions_pdf():
    # Create temporary file
    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.pdf')
    temp_file.close()
    
    doc = SimpleDocTemplate(temp_file.name, pagesize=letter)
    styles = getSampleStyleSheet()
    story = []
    
    # Title
    title_style = styles['Title']
    story.append(Paragraph("Organic Reactions Database", title_style))
    story.append(Spacer(1, 12))
    
    for idx, row in df.iterrows():
        # Reaction header
        reaction_title = f"Reaction {idx+1}: {row['corrected_name']}"
        story.append(Paragraph(reaction_title, styles['Heading2']))
        
        # Generate SVG for this reaction
        try:
            reactants_smiles = '.'.join([s for s in row['reactants_smiles'] if s is not None])
            products_smiles = '.'.join([s for s in row['products_smiles'] if s is not None])
            if reactants_smiles and products_smiles:
                reaction_smiles = f"{reactants_smiles}>>{products_smiles}"
                rxn = AllChem.ReactionFromSmarts(reaction_smiles)
                if rxn:
                    svg_content = Draw.ReactionToImage(rxn, useSVG=True)
                    # Save SVG to temp file
                    svg_temp = tempfile.NamedTemporaryFile(delete=False, suffix='.svg')
                    svg_temp.write(svg_content.encode('utf-8'))
                    svg_temp.close()
                    # Convert SVG to ReportLab drawing
                    drawing = svg2rlg(svg_temp.name)
                    if drawing:
                        # Scale the drawing to fit
                        drawing.width = 400
                        drawing.height = 150
                        drawing.scale(0.8, 0.8)
                        story.append(drawing)
                        story.append(Spacer(1, 12))
                    # Clean up temp file
                    os.unlink(svg_temp.name)
        except Exception as e:
            # If SVG generation fails, just continue
            pass
        
        # Handle potential None values in SMILES
        reactants_smiles = [s for s in row['reactants_smiles'] if s is not None]
        reagents_smiles = [s for s in row['reagents_smiles'] if s is not None]
        products_smiles = [s for s in row['products_smiles'] if s is not None]
        
        # Content
        reactants = row['general_reactants'] if pd.notna(row['general_reactants']) else 'N/A'
        reagents = row['general_reagents'] if pd.notna(row['general_reagents']) else 'N/A'
        products = row['general_products'] if pd.notna(row['general_products']) else 'N/A'
        
        content = [
            f"<b>Reactants:</b> {reactants}",
            f"<b>Reactants SMILES:</b> {', '.join(reactants_smiles) if reactants_smiles else 'None'}",
            f"<b>Reagents:</b> {reagents}",
            f"<b>Reagents SMILES:</b> {', '.join(reagents_smiles) if reagents_smiles else 'None'}",
            f"<b>Products:</b> {products}",
            f"<b>Products SMILES:</b> {', '.join(products_smiles) if products_smiles else 'None'}"
        ]
        
        for item in content:
            story.append(Paragraph(item, styles['Normal']))
        
        story.append(Spacer(1, 12))
    
    doc.build(story)
    return temp_file.name

def search_by_reaction_name(query):
    if not query:
        return "Please enter a reaction name."
    # Exact match first
    result = df[df['corrected_name'].str.lower() == query.lower()]
    if not result.empty:
        row = result.iloc[0]
        reactants = row['general_reactants'] if pd.notna(row['general_reactants']) else 'N/A'
        products = row['general_products'] if pd.notna(row['general_products']) else 'N/A'
        reagents = row['general_reagents'] if pd.notna(row['general_reagents']) else 'N/A'
        return f"**{row['corrected_name']}**\n\n**Reactants:** {reactants}\n\n**Reagents:** {reagents}\n\n**Products:** {products}\n\n**Description:** {row['description'][:500]}..."
    # Fuzzy match
    matches = process.extract(query, reaction_names, limit=1)
    if matches and matches[0][1] > 80:
        best_match = matches[0][0]
        result = df[df['corrected_name'] == best_match]
        row = result.iloc[0]
        reactants = row['general_reactants'] if pd.notna(row['general_reactants']) else 'N/A'
        products = row['general_products'] if pd.notna(row['general_products']) else 'N/A'
        reagents = row['general_reagents'] if pd.notna(row['general_reagents']) else 'N/A'
        return f"**{row['corrected_name']}** (closest match)\n\n**Reactants:** {reactants}\n\n**Reagents:** {reagents}\n\n**Products:** {products}\n\n**Description:** {row['description'][:500]}..."
    return "No matching reaction found."

def search_by_reactant(reactant):
    if not reactant:
        return "Please enter a reactant."
    matches = df[df['general_reactants'].str.lower().str.contains(reactant.lower(), na=False)]
    if not matches.empty:
        results = []
        for _, row in matches.head(5).iterrows():
            reactants = row['general_reactants'] if pd.notna(row['general_reactants']) else 'N/A'
            products = row['general_products'] if pd.notna(row['general_products']) else 'N/A'
            results.append(f"**{row['corrected_name']}**: {reactants} → {products}")
        return "\n\n".join(results)
    # Fuzzy match for autocorrection
    fuzzy_matches = process.extract(reactant, unique_reactants, limit=3)
    if fuzzy_matches and fuzzy_matches[0][1] > 70:
        closest = fuzzy_matches[0][0]
        matches = df[df['general_reactants'].str.lower().str.contains(closest.lower(), na=False)]
        if not matches.empty:
            results = [f"Did you mean '{closest}'?\n"]
            for _, row in matches.head(5).iterrows():
                reactants = row['general_reactants'] if pd.notna(row['general_reactants']) else 'N/A'
                products = row['general_products'] if pd.notna(row['general_products']) else 'N/A'
                results.append(f"**{row['corrected_name']}**: {reactants} → {products}")
            return "\n\n".join(results)
    return "No reactions found with that reactant."

def search_by_product(product):
    if not product:
        return "Please enter a product."
    matches = df[df['general_products'].str.lower().str.contains(product.lower(), na=False)]
    if not matches.empty:
        results = []
        for _, row in matches.head(5).iterrows():
            reactants = row['general_reactants'] if pd.notna(row['general_reactants']) else 'N/A'
            products = row['general_products'] if pd.notna(row['general_products']) else 'N/A'
            results.append(f"**{row['corrected_name']}**: {reactants} → {products}")
        return "\n\n".join(results)
    # Fuzzy match for autocorrection
    fuzzy_matches = process.extract(product, unique_products, limit=3)
    if fuzzy_matches and fuzzy_matches[0][1] > 70:
        closest = fuzzy_matches[0][0]
        matches = df[df['general_products'].str.lower().str.contains(closest.lower(), na=False)]
        if not matches.empty:
            results = [f"Did you mean '{closest}'?\n"]
            for _, row in matches.head(5).iterrows():
                reactants = row['general_reactants'] if pd.notna(row['general_reactants']) else 'N/A'
                products = row['general_products'] if pd.notna(row['general_products']) else 'N/A'
                results.append(f"**{row['corrected_name']}**: {reactants} → {products}")
            return "\n\n".join(results)
    return "No reactions found with that product."

def get_autocomplete_reactions(query):
    if not query:
        return reaction_names[:10]
    matches = process.extract(query, reaction_names, limit=10)
    return [m[0] for m in matches if m[1] > 60]

def fix_reaction_with_gemini(reaction_name, api_key):
    if not api_key:
        return "Please provide a Gemini API key."
    
    try:
        # Find the reaction row index
        result = df[df['corrected_name'].str.lower() == reaction_name.lower()]
        if result.empty:
            return f"❌ Reaction '{reaction_name}' not found in database."
        row_index = result.index[0]
        
        client = genai.Client(api_key=api_key)
        
        prompt = f"""Please provide detailed information about the organic reaction named "{reaction_name}". 
        Include the correct reaction name, reactants, reagents, products, byproducts, reaction conditions, mechanism, and description.
        Make sure to provide accurate chemical information."""
        
        contents = [
            types.Content(
                role="user",
                parts=[types.Part.from_text(text=prompt)],
            ),
        ]
        
        generate_content_config = types.GenerateContentConfig(
            thinking_config=types.ThinkingConfig(thinking_budget=-1),
            response_mime_type="application/json",
            response_schema=genai.types.Schema(
                type=genai.types.Type.OBJECT,
                required=["reaction name", "reactants", "reagents", "products", "byproducts", "conditions", "mechanism", "description"],
                properties={
                    "reaction name": genai.types.Schema(type=genai.types.Type.STRING),
                    "reactants": genai.types.Schema(
                        type=genai.types.Type.ARRAY,
                        items=genai.types.Schema(type=genai.types.Type.STRING),
                    ),
                    "reagents": genai.types.Schema(
                        type=genai.types.Type.ARRAY,
                        items=genai.types.Schema(type=genai.types.Type.STRING),
                    ),
                    "products": genai.types.Schema(
                        type=genai.types.Type.ARRAY,
                        items=genai.types.Schema(type=genai.types.Type.STRING),
                    ),
                    "byproducts": genai.types.Schema(
                        type=genai.types.Type.ARRAY,
                        items=genai.types.Schema(type=genai.types.Type.STRING),
                    ),
                    "conditions": genai.types.Schema(type=genai.types.Type.STRING),
                    "mechanism": genai.types.Schema(type=genai.types.Type.STRING),
                    "description": genai.types.Schema(type=genai.types.Type.STRING),
                },
            ),
        )
        
        response_text = ""
        for chunk in client.models.generate_content_stream(
            model="gemini-2.5-pro",
            contents=contents,
            config=generate_content_config,
        ):
            response_text += chunk.text
        
        # Parse the JSON response
        import json
        gemini_data = json.loads(response_text)
        
        # Store the updated data globally for potential database update
        global last_ai_fix
        last_ai_fix = {
            'reaction_name': reaction_name,
            'row_index': row_index,
            'updated_data': gemini_data,
            'timestamp': str(pd.Timestamp.now())
        }
        
        return f"✅ **AI Fix Completed for '{reaction_name}'**\n\n**Updated Data:**\n- **Name:** {gemini_data.get('reaction name', 'N/A')}\n- **Reactants:** {', '.join(gemini_data.get('reactants', []))}\n- **Reagents:** {', '.join(gemini_data.get('reagents', []))}\n- **Products:** {', '.join(gemini_data.get('products', []))}\n- **Description:** {gemini_data.get('description', '')[:200]}...\n\n💡 **To save this fix to the database, enter the admin password below and click 'Update Database'.**"
            
    except Exception as e:
        return f"❌ Error calling Gemini API: {str(e)}"

def update_database_with_ai_fix(password):
    if password != "Vvn@#411037":
        return "❌ Incorrect password. Database update denied."
    
    global last_ai_fix
    if not last_ai_fix:
        return "❌ No recent AI fix to save. Please fix a reaction first."
    
    try:
        # Update the global df using the stored row index
        global df
        idx = last_ai_fix['row_index']
        
        # Verify the row still exists
        if idx not in df.index:
            return f"❌ Row index {idx} not found in database. The data may have been modified."
        
        # Store original values for logging
        original_name = df.at[idx, 'corrected_name']
        
        # Update the dataframe
        df.at[idx, 'corrected_name'] = last_ai_fix['updated_data'].get('reaction name', last_ai_fix['reaction_name'])
        df.at[idx, 'general_reactants'] = ', '.join(last_ai_fix['updated_data'].get('reactants', []))
        df.at[idx, 'general_reagents'] = ', '.join(last_ai_fix['updated_data'].get('reagents', []))
        df.at[idx, 'general_products'] = ', '.join(last_ai_fix['updated_data'].get('products', []))
        df.at[idx, 'description'] = last_ai_fix['updated_data'].get('description', df.at[idx, 'description'])
        
        # Try to push to Hugging Face
        hf_token = os.getenv('trial1')
        if hf_token:
            try:
                # Convert back to Hugging Face dataset
                updated_dataset = Dataset.from_pandas(df)
                
                # Initialize HF API
                api = HfApi()
                
                # Push to Hugging Face
                updated_dataset.push_to_hub(
                    "smitathkr1/organic_reactions_corrected",
                    token=hf_token,
                    commit_message=f"AI fix: Updated reaction '{original_name}' -> '{df.at[idx, 'corrected_name']}'"
                )
                
                push_success = True
            except Exception as push_error:
                push_success = False
                push_error_msg = str(push_error)
        else:
            push_success = False
            push_error_msg = "trial1 not found in environment variables"
        
        # Log the update
        log_entry = f"[{last_ai_fix['timestamp']}] Database updated: '{original_name}' -> '{df.at[idx, 'corrected_name']}' | HF Push: {'Success' if push_success else 'Failed: ' + push_error_msg}\n"
        with open('database_updates.log', 'a') as f:
            f.write(log_entry)
        
        # Update the global reaction_names list in case the name changed
        global reaction_names
        reaction_names = df['corrected_name'].unique().tolist()
        
        # Clear the last fix
        last_ai_fix = None
        
        success_msg = "✅ **Database Updated Successfully!**\n\n"
        if push_success:
            success_msg += "The reaction has been permanently updated on Hugging Face and is now live!\n\n"
        else:
            success_msg += "The reaction has been updated in the current session.\n"
            success_msg += f"**Note:** Could not push to Hugging Face: {push_error_msg}\n"
            success_msg += "Please check that trial1 is set in space secrets.\n\n"
        
        success_msg += "Changes logged to 'database_updates.log'."
        
        return success_msg
            
    except Exception as e:
        return f"❌ Error updating database: {str(e)}"

def get_autocomplete_reactants(query):
    if not query:
        return unique_reactants[:10]
    matches = process.extract(query, unique_reactants, limit=10)
    return [m[0] for m in matches if m[1] > 60]

def get_autocomplete_products(query):
    if not query:
        return unique_products[:10]
    matches = process.extract(query, unique_products, limit=10)
    return [m[0] for m in matches if m[1] > 60]

with gr.Blocks(title="Organic Reactions Search") as demo:
    gr.Markdown("# Organic Reactions Search API")
    gr.Markdown("Search through the organic reactions dataset by name, reactant, or product.")

    with gr.Tab("Search by Reaction Name"):
        reaction_input = gr.Dropdown(label="Reaction Name", choices=reaction_names, allow_custom_value=True)
        reaction_output = gr.Markdown(label="Result")
        reaction_btn = gr.Button("Search")
        reaction_btn.click(search_by_reaction_name, inputs=reaction_input, outputs=reaction_output)

    with gr.Tab("Search by Reactant"):
        reactant_input = gr.Dropdown(label="Reactant", choices=unique_reactants, allow_custom_value=True)
        reactant_output = gr.Markdown(label="Results")
        reactant_btn = gr.Button("Search")
        reactant_btn.click(search_by_reactant, inputs=reactant_input, outputs=reactant_output)

    with gr.Tab("View Reaction SVG"):
        svg_input = gr.Dropdown(label="Reaction Name", choices=reaction_names, allow_custom_value=True)
        svg_output = gr.HTML(label="Reaction SVG")
        svg_btn = gr.Button("Generate SVG")
        svg_btn.click(generate_reaction_svg, inputs=svg_input, outputs=svg_output)

    with gr.Tab("Search by Product"):
        product_input = gr.Dropdown(label="Product", choices=unique_products, allow_custom_value=True)
        product_output = gr.Markdown(label="Results")
        product_btn = gr.Button("Search")
        product_btn.click(search_by_product, inputs=product_input, outputs=product_output)

    with gr.Tab("Download All Reactions PDF"):
        gr.Markdown("Download a comprehensive PDF containing all 828 reactions with their names, reactants, reagents, products, and SMILES strings.")
        pdf_btn = gr.Button("Generate and Download PDF")
        pdf_output = gr.File(label="Download PDF")
        pdf_btn.click(generate_all_reactions_pdf, outputs=pdf_output)

    with gr.Tab("View All Reactions (Table)"):
        gr.Markdown("Browse all 828 reactions in a tabular format. Use the AI Fix section below to improve reaction data.")
        
        # AI Fix section
        with gr.Row():
            api_key_input = gr.Textbox(label="Gemini API Key", type="password", placeholder="Enter your Gemini API key")
            reaction_to_fix = gr.Dropdown(label="Select Reaction to Fix", choices=reaction_names)
        
        fix_button = gr.Button("Fix with AI")
        ai_status = gr.Markdown(label="AI Fix Status")
        
        fix_button.click(fix_reaction_with_gemini, inputs=[reaction_to_fix, api_key_input], outputs=ai_status)
        
        # Database update section
        gr.Markdown("---")
        gr.Markdown("**Database Update (Admin Only):**")
        with gr.Row():
            admin_password = gr.Textbox(label="Admin Password", type="password", placeholder="Enter admin password to update database")
            update_db_button = gr.Button("Update Database", variant="secondary")
        
        update_status = gr.Markdown(label="Update Status")
        update_db_button.click(update_database_with_ai_fix, inputs=[admin_password], outputs=update_status)
        
        gr.Markdown("---")
        gr.Markdown("**Database Table:**")
        
        # Create HTML table (read-only for browsing)
        def create_reactions_table():
            html = """
            <table style="width:100%; border-collapse: collapse;">
                <thead>
                    <tr style="background-color: #f2f2f2;">
                        <th style="border: 1px solid #ddd; padding: 8px;">Reaction Name</th>
                        <th style="border: 1px solid #ddd; padding: 8px;">Reactants</th>
                        <th style="border: 1px solid #ddd; padding: 8px;">Reactants SMILES</th>
                        <th style="border: 1px solid #ddd; padding: 8px;">Reagents</th>
                        <th style="border: 1px solid #ddd; padding: 8px;">Reagents SMILES</th>
                        <th style="border: 1px solid #ddd; padding: 8px;">Products</th>
                        <th style="border: 1px solid #ddd; padding: 8px;">Products SMILES</th>
                        <th style="border: 1px solid #ddd; padding: 8px;">Description</th>
                    </tr>
                </thead>
                <tbody>
            """
            
            for idx, row in df.iterrows():
                reaction_name = row['corrected_name']
                reactants = row['general_reactants'] if pd.notna(row['general_reactants']) else 'N/A'
                reactants_smiles_list = [s for s in row['reactants_smiles'] if s is not None and pd.notna(s)]
                reactants_smiles = ', '.join(reactants_smiles_list) if reactants_smiles_list else 'N/A'
                reagents = row['general_reagents'] if pd.notna(row['general_reagents']) else 'N/A'
                reagents_smiles_list = [s for s in row['reagents_smiles'] if s is not None and pd.notna(s)]
                reagents_smiles = ', '.join(reagents_smiles_list) if reagents_smiles_list else 'N/A'
                products = row['general_products'] if pd.notna(row['general_products']) else 'N/A'
                products_smiles_list = [s for s in row['products_smiles'] if s is not None and pd.notna(s)]
                products_smiles = ', '.join(products_smiles_list) if products_smiles_list else 'N/A'
                description = row['description'][:100] + '...' if len(row['description']) > 100 else row['description']
                
                html += f"""
                    <tr>
                        <td style="border: 1px solid #ddd; padding: 8px;">{reaction_name}</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{reactants}</td>
                        <td style="border: 1px solid #ddd; padding: 8px; font-family: monospace; font-size: 12px;">{reactants_smiles}</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{reagents}</td>
                        <td style="border: 1px solid #ddd; padding: 8px; font-family: monospace; font-size: 12px;">{reagents_smiles}</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{products}</td>
                        <td style="border: 1px solid #ddd; padding: 8px; font-family: monospace; font-size: 12px;">{products_smiles}</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{description}</td>
                    </tr>
                """
            
            html += "</tbody></table>"
            return html
        
        table_html = create_reactions_table()
        table_output = gr.HTML(value=table_html, label="All Reactions Database")
        gr.Markdown("""
        ## API Endpoints
        
        This Gradio app exposes the following functions as API endpoints. You can call them via HTTP POST requests to the `/api/predict` endpoint.
        
        ### Search by Reaction Name (fn_index: 0)
        - **Input**: `query` (string) - The reaction name to search
        - **Output**: Markdown string with reaction details
        
        ### Search by Reactant (fn_index: 1)
        - **Input**: `reactant` (string) - The reactant to search for
        - **Output**: Markdown string with matching reactions
        
        ### Search by Product (fn_index: 2)
        - **Input**: `product` (string) - The product to search for
        - **Output**: Markdown string with matching reactions
        
        ### Autocomplete Reaction Names (fn_index: 3)
        - **Input**: `query` (string) - Partial reaction name
        - **Output**: List of matching reaction names
        
        ### Autocomplete Reactants (fn_index: 4)
        - **Input**: `query` (string) - Partial reactant name
        - **Output**: List of matching reactants
        
        ### Autocomplete Products (fn_index: 5)
        - **Input**: `query` (string) - Partial product name
        - **Output**: List of matching products
        
        ### Generate Reaction SVG (fn_index: 6)
        - **Input**: `name` (string) - Exact reaction name
        - **Output**: SVG string of the reaction diagram
        
        ### Generate All Reactions PDF (fn_index: 7)
        - **Input**: None
        - **Output**: PDF file with all 828 reactions data
        
        ### Example API Call
        ```bash
        curl -X POST "https://smitathkr1-namereaction-api.hf.space/api/predict" \\
             -H "Content-Type: application/json" \\
             -d '{"fn_index": 0, "data": ["appel-reaction"]}'
        ```
        
        Note: `fn_index` corresponds to the function order in the app (0-based).
        """)

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