app.py

# --- IMPORTS ---
# Core and Data Handling
import gradio as gr
import pandas as pd
import numpy as np
import os
import glob
import time
import warnings

# Chemistry and Cheminformatics
from rdkit import Chem
from rdkit.Chem import Descriptors, Lipinski
from chembl_webresource_client.new_client import new_client
from padelpy import padeldescriptor
# Removed: import mols2grid
from rdkit.Chem.Draw import rdMolDraw2D
from rdkit.Chem import Draw
import base64
from io import BytesIO


# Plotting and Visualization
import matplotlib.pyplot as plt
import seaborn as sns
from scipy import stats
from scipy.stats import mannwhitneyu

# Machine Learning Models and Metrics
from sklearn.model_selection import train_test_split
from sklearn.feature_selection import VarianceThreshold
from sklearn.linear_model import (
    LinearRegression, Ridge, Lasso, ElasticNet, BayesianRidge,
    HuberRegressor, PassiveAggressiveRegressor, OrthogonalMatchingPursuit,
    LassoLars
)
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import (
    RandomForestRegressor, GradientBoostingRegressor, ExtraTreesRegressor,
    AdaBoostRegressor
)
from sklearn.neighbors import KNeighborsRegressor
from sklearn.dummy import DummyRegressor
from sklearn.metrics import (
    mean_absolute_error, mean_squared_error, r2_score
)

# A placeholder class to store all results from a modeling run
class ModelRunResult:
    def __init__(self, dataframe, plotter, models, selected_features):
        self.dataframe = dataframe
        self.plotter = plotter
        self.models = models
        self.selected_features = selected_features

# Optional Advanced Models
try:
    import xgboost as xgb
    import lightgbm as lgb
    import catboost as cb
    _has_extra_libs = True
except ImportError:
    _has_extra_libs = False
    warnings.warn("Optional libraries (xgboost, lightgbm, catboost) not found. Some models will be unavailable.")

# --- GLOBAL CONFIGURATION & SETUP ---
warnings.filterwarnings("ignore")
sns.set_theme(style='whitegrid')

# --- FINGERPRINT CONFIGURATION ---
# Create a dummy PubChem.xml if no XML files are found, to ensure fp_config is populated
# Updated path for XML files to 'padel_descriptors/*.xml'
padel_descriptors_dir = 'padel_descriptors'
if not os.path.exists(padel_descriptors_dir):
    os.makedirs(padel_descriptors_dir)

# Check for XML files within the 'padel_descriptors' folder
xml_files = sorted(glob.glob(os.path.join(padel_descriptors_dir, '*.xml')))
if not xml_files:
    try:
        # Create a dummy PubChem.xml inside 'padel_descriptors' if no XML files are found
        with open(os.path.join(padel_descriptors_dir, 'PubChem.xml'), 'w') as f:
            f.write('')
        xml_files = sorted(glob.glob(os.path.join(padel_descriptors_dir, '*.xml'))) # Re-scan after creating dummy
    except IOError:
        warnings.warn("Could not create a dummy 'PubChem.xml' file in 'padel_descriptors'. Fingerprint calculation might fail if no .xml files are present.")

if not xml_files:
    warnings.warn(
        "No descriptor .xml files found in 'padel_descriptors' directory. "
        "Fingerprint calculation will not be possible. "
        "Please place descriptor XML files in the 'padel_descriptors' directory."
    )
fp_config = {os.path.splitext(os.path.basename(file))[0]: file for file in xml_files}
FP_list = sorted(list(fp_config.keys()))


# ==============================================================================
# === STEP 1: CORE DATA COLLECTION & EDA FUNCTIONS ===
# ==============================================================================

def get_target_chembl_id(query):
    try:
        target = new_client.target
        res = target.search(query)
        if not res:
            return pd.DataFrame(), gr.Dropdown(choices=[], value=None), "No targets found for your query."
        df = pd.DataFrame(res)
        return df[["target_chembl_id", "pref_name", "organism"]], gr.Dropdown(choices=df["target_chembl_id"].tolist()), f"Found {len(df)} targets."
    except Exception as e:
        raise gr.Error(f"ChEMBL search failed: {e}")

def get_bioactivity_data(target_id):
    try:
        activity = new_client.activity
        res = activity.filter(target_chembl_id=target_id).filter(standard_type="IC50")
        if not res:
            return pd.DataFrame(), "No IC50 bioactivity data found for this target."
        df = pd.DataFrame(res)
        return df, f"Fetched {len(df)} data points."
    except Exception as e:
        raise gr.Error(f"Failed to fetch bioactivity data: {e}")

def pIC50_calc(input_df):
    df_copy = input_df.copy()
    df_copy['standard_value'] = pd.to_numeric(df_copy['standard_value'], errors='coerce')
    df_copy.dropna(subset=['standard_value'], inplace=True)
    df_copy['standard_value_norm'] = df_copy['standard_value'].apply(lambda x: min(x, 100000000))
    pIC50_values = []
    for i in df_copy['standard_value_norm']:
        if pd.notna(i) and i > 0:
            molar = i * (10**-9)
            pIC50_values.append(-np.log10(molar))
        else:
            pIC50_values.append(np.nan)
    df_copy['pIC50'] = pIC50_values
    df_copy['bioactivity_class'] = df_copy['standard_value_norm'].apply(
        lambda x: "inactive" if pd.notna(x) and x >= 10000 else ("active" if pd.notna(x) and x <= 1000 else "intermediate")
    )
    return df_copy.drop(columns=['standard_value', 'standard_value_norm'])

def lipinski_descriptors(smiles_series):
    moldata, valid_smiles = [], []
    for elem in smiles_series:
        if elem and isinstance(elem, str):
            mol = Chem.MolFromSmiles(elem)
            if mol:
                moldata.append(mol)
                valid_smiles.append(elem)
    descriptor_rows = []
    for mol in moldata:
        row = [Descriptors.MolWt(mol), Descriptors.MolLogP(mol), Lipinski.NumHDonors(mol), Lipinski.NumHAcceptors(mol)]
        descriptor_rows.append(row)
    columnNames = ["MW", "LogP", "NumHDonors", "NumHAcceptors"]
    if not descriptor_rows: return pd.DataFrame(columns=columnNames), []
    return pd.DataFrame(data=np.array(descriptor_rows), columns=columnNames), valid_smiles

def clean_and_process_data(df):
    if df is None or df.empty: raise gr.Error("No data to process. Please fetch data first.")
    if "canonical_smiles" not in df.columns or df["canonical_smiles"].isnull().all():
        try:
            df["canonical_smiles"] = [c.get("molecule_structures", {}).get("canonical_smiles") for c in new_client.molecule.get(list(df["molecule_chembl_id"]))]
        except Exception as e:
            raise gr.Error(f"Could not fetch SMILES from ChEMBL: {e}")
    df = df[df.standard_value.notna()]
    df = df[df.canonical_smiles.notna()]
    # DEBUG FIX: Added drop_duplicates to align with notebook logic and ensure unique SMILES for merging.
    df.drop_duplicates(['canonical_smiles'], inplace=True)
    df["standard_value"] = pd.to_numeric(df["standard_value"], errors='coerce')
    df.dropna(subset=['standard_value'], inplace=True)
    df_processed = pIC50_calc(df)
    df_processed = df_processed[df_processed.pIC50.notna()]
    if df_processed.empty: return pd.DataFrame(), "No compounds remaining after pIC50 calculation."
    df_lipinski, valid_smiles = lipinski_descriptors(df_processed['canonical_smiles'])
    if not valid_smiles: return pd.DataFrame(), "No valid SMILES could be processed for Lipinski descriptors."
    df_processed = df_processed[df_processed['canonical_smiles'].isin(valid_smiles)].reset_index(drop=True)
    df_lipinski = df_lipinski.reset_index(drop=True)
    df_final = pd.concat([df_processed, df_lipinski], axis=1)
    return df_final, f"Processing complete. {len(df_final)} compounds remain after cleaning."

def run_eda_analysis(df, selected_classes):
    if df is None or df.empty: raise gr.Error("No data available for analysis.")
    df_filtered = df[df.bioactivity_class.isin(selected_classes)].copy()
    if df_filtered.empty: return (None, None, None, pd.DataFrame(), None, pd.DataFrame(), None, pd.DataFrame(), None, pd.DataFrame(), None, pd.DataFrame(), "No data for selected classes.")
    plots = [create_frequency_plot(df_filtered), create_scatter_plot(df_filtered)]
    stats_dfs = []
    for desc in ['pIC50', 'MW', 'LogP', 'NumHDonors', 'NumHAcceptors']:
        plots.append(create_boxplot(df_filtered, desc))
        stats_dfs.append(mannwhitney_test(df_filtered, desc))
    plt.close('all')
    return (plots[0], plots[1], plots[2], stats_dfs[0], plots[3], stats_dfs[1], plots[4], stats_dfs[2], plots[5], stats_dfs[3], plots[6], stats_dfs[4], f"EDA complete for {len(df_filtered)} compounds.")

def create_frequency_plot(df):
    plt.figure(figsize=(5.5, 5.5)); sns.barplot(x=df['bioactivity_class'].value_counts().index, y=df['bioactivity_class'].value_counts().values, palette={'active': '#1f77b4', 'inactive': '#ff7f0e', 'intermediate': '#2ca02c'}); plt.xlabel('Bioactivity Class', fontsize=12); plt.ylabel('Frequency', fontsize=12); plt.title('Frequency of Bioactivity Classes', fontsize=14); return plt.gcf()
def create_scatter_plot(df):
    plt.figure(figsize=(5.5, 5.5)); sns.scatterplot(data=df, x='MW', y='LogP', hue='bioactivity_class', size='pIC50', palette={'active': '#1f77b4', 'inactive': '#ff7f0e', 'intermediate': '#2ca02c'}, sizes=(20, 200), alpha=0.7); plt.xlabel('Molecular Weight (MW)', fontsize=12); plt.ylabel('LogP', fontsize=12); plt.title('Chemical Space: MW vs. LogP', fontsize=14); plt.legend(title='Bioactivity Class'); return plt.gcf()
def create_boxplot(df, descriptor):
    plt.figure(figsize=(5.5, 5.5)); sns.boxplot(x='bioactivity_class', y=descriptor, data=df, palette={'active': '#1f77b4', 'inactive': '#ff7f0e', 'intermediate': '#2ca02c'}); plt.xlabel('Bioactivity Class', fontsize=12); plt.ylabel(descriptor, fontsize=12); plt.title(f'{descriptor} by Bioactivity Class', fontsize=14); return plt.gcf()
def mannwhitney_test(df, descriptor):
    results = []
    for c1, c2 in [('active', 'inactive'), ('active', 'intermediate'), ('inactive', 'intermediate')]:
        if c1 in df['bioactivity_class'].unique() and c2 in df['bioactivity_class'].unique():
            d1, d2 = df[df.bioactivity_class == c1][descriptor].dropna(), df[df.bioactivity_class == c2][descriptor].dropna()
            if not d1.empty and not d2.empty:
                stat, p = mannwhitneyu(d1, d2)
                results.append({'Comparison': f'{c1.title()} vs {c2.title()}', 'Statistics': stat, 'p-value': p, 'Interpretation': 'Different distribution (p < 0.05)' if p <= 0.05 else 'Same distribution (p > 0.05)'})
    return pd.DataFrame(results)

# ==============================================================================
# === STEP 2: FEATURE ENGINEERING FUNCTIONS ===
# ==============================================================================
# Replacement for mols2grid.display in Step 2
def create_molecule_grid_html(df, smiles_col='canonical_smiles', max_mols=20):
    html_parts = ['<div style="display: flex; flex-wrap: wrap; gap: 10px;">']
    for idx, row in df.head(max_mols).iterrows():
        smiles = row[smiles_col]
        pic50 = row['pIC50']
        mol = Chem.MolFromSmiles(smiles)
        if mol:
            # Generate molecule image
            img = Draw.MolToImage(mol, size=(200, 200))
            # Convert to base64
            buffered = BytesIO()
            img.save(buffered, format="PNG")
            img_str = base64.b64encode(buffered.getvalue()).decode()
            # Create HTML for this molecule
            mol_html = f'''
            <div style="border: 1px solid #ccc; padding: 10px; border-radius: 5px; text-align: center;">
                <img src="data:image/png;base64,{img_str}" alt="Molecule" style="max-width: 200px;">
                <p><strong>pIC50:</strong> {pic50:.2f}</p>
                <p style="font-size: 10px; word-break: break-all;">{smiles}</p>
            </div>
            '''
            html_parts.append(mol_html)
    html_parts.append('</div>')
    return ''.join(html_parts)

def calculate_fingerprints(current_state, fingerprint_type, progress=gr.Progress()):
    input_df = current_state.get('cleaned_data')
    if input_df is None or input_df.empty: raise gr.Error("No cleaned data found. Please complete Step 1.")
    if not fingerprint_type: raise gr.Error("Please select a fingerprint type.")
    progress(0, desc="Starting..."); yield f"🧪 Starting fingerprint calculation...", None, gr.update(visible=False), None, current_state
    try:
        smi_file, output_csv = 'molecule.smi', 'fingerprints.csv'
        
        # DEBUG FIX: Switched to a safe merge instead of risky concat.
        # Use canonical_smiles as the unique ID for PaDEL, since it was deduplicated in Step 1.
        input_df[['canonical_smiles', 'canonical_smiles']].to_csv(smi_file, sep='\t', index=False, header=False)
        
        if os.path.exists(output_csv): os.remove(output_csv)
        descriptortypes = fp_config.get(fingerprint_type)
        if not descriptortypes: raise gr.Error(f"Descriptor XML for '{fingerprint_type}' not found.")
        
        progress(0.3, desc="⚗️ Running PaDEL..."); yield f"⚗️ Running PaDEL...", None, gr.update(visible=False), None, current_state
        padeldescriptor(mol_dir=smi_file, d_file=output_csv, descriptortypes=descriptortypes, detectaromaticity=True, standardizenitro=True, standardizetautomers=True, threads=-1, removesalt=True, log=False, fingerprints=True)
        if not os.path.exists(output_csv) or os.path.getsize(output_csv) == 0:
            raise gr.Error("PaDEL failed to produce an output file. Check molecule validity.")

        progress(0.7, desc="📊 Processing results..."); yield "📊 Processing results...", None, gr.update(visible=False), None, current_state
        df_X = pd.read_csv(output_csv).rename(columns={'Name': 'canonical_smiles'})
        
        # Safely merge fingerprints with original data. 'inner' ensures that only molecules
        # for which fingerprints were successfully calculated are included.
        final_df = pd.merge(input_df[['canonical_smiles', 'pIC50']], df_X, on='canonical_smiles', how='inner')
        
        current_state['fingerprint_data'] = final_df; current_state['fingerprint_type'] = fingerprint_type
        progress(0.9, desc="🖼️ Generating molecule grid...")
        # Replacement for mols2grid.display in Step 2
        mols_html = create_molecule_grid_html(final_df)
        success_msg = f"✅ Success! Generated {len(df_X.columns) -1} descriptors for {len(final_df)} molecules."
        progress(1, desc="Completed!"); yield success_msg, final_df, gr.update(visible=True), gr.update(value=mols_html, visible=True), current_state
    except Exception as e: raise gr.Error(f"Calculation failed: {e}")
    finally:
        if os.path.exists('molecule.smi'): os.remove('molecule.smi')
        if os.path.exists('fingerprints.csv'): os.remove('fingerprints.csv')

# ==============================================================================
# === STEP 3: MODEL TRAINING & PREDICTION FUNCTIONS ===
# ==============================================================================
class ModelPlotter:
    def __init__(self, models: dict, X_test: pd.DataFrame, y_test: pd.Series):
        self._models, self._X_test, self._y_test = models, X_test, y_test
    def plot_validation(self, model_name: str):
        if model_name not in self._models: raise ValueError(f"Model '{model_name}' not found.")
        model, y_pred = self._models[model_name], self._models[model_name].predict(self._X_test)
        residuals = self._y_test - y_pred
        fig, axes = plt.subplots(2, 2, figsize=(12, 10)); fig.suptitle(f'Model Validation Plots for {model_name}', fontsize=16, y=1.02)
        sns.scatterplot(x=self._y_test, y=y_pred, ax=axes[0, 0], alpha=0.6); axes[0, 0].set_title('Actual vs. Predicted'); axes[0, 0].set_xlabel('Actual pIC50'); axes[0, 0].set_ylabel('Predicted pIC50'); lims = [min(self._y_test.min(), y_pred.min()), max(self._y_test.max(), y_pred.max())]; axes[0, 0].plot(lims, lims, 'r--', alpha=0.75, zorder=0)
        sns.scatterplot(x=y_pred, y=residuals, ax=axes[0, 1], alpha=0.6); axes[0, 1].axhline(y=0, color='r', linestyle='--'); axes[0, 1].set_title('Residuals vs. Predicted'); axes[0, 1].set_xlabel('Predicted pIC50'); axes[0, 1].set_ylabel('Residuals')
        sns.histplot(residuals, kde=True, ax=axes[1, 0]); axes[1, 0].set_title('Distribution of Residuals')
        stats.probplot(residuals, dist="norm", plot=axes[1, 1]); axes[1, 1].set_title('Normal Q-Q Plot')
        plt.tight_layout(); return fig
    def plot_feature_importance(self, model_name: str, top_n: int = 7):
        if model_name not in self._models: raise ValueError(f"Model '{model_name}' not found.")
        model = self._models[model_name]
        if hasattr(model, 'feature_importances_'): importances = model.feature_importances_
        elif hasattr(model, 'coef_'): importances = np.abs(model.coef_)
        else: return None
        top_features = pd.DataFrame({'Feature': self._X_test.columns, 'Importance': importances}).sort_values(by='Importance', ascending=False).head(top_n)
        plt.figure(figsize=(10, top_n * 0.5)); sns.barplot(x='Importance', y='Feature', data=top_features, palette='viridis', orient='h'); plt.title(f'Top {top_n} Features for {model_name}'); plt.tight_layout(); return plt.gcf()

def run_regression_suite(df: pd.DataFrame, progress=gr.Progress()):
    progress(0, desc="Splitting data..."); yield "Splitting data (80/20 train/test split)...", None, None
    X = df.drop(columns=['pIC50', 'canonical_smiles'], errors='ignore')
    y = df['pIC50']
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

    progress(0.1, desc="Selecting features..."); yield "Performing feature selection (removing low variance)...", None, None
    selector = VarianceThreshold(threshold=0.1)
    X_train = pd.DataFrame(selector.fit_transform(X_train), columns=X_train.columns[selector.get_support()], index=X_train.index)
    X_test = pd.DataFrame(selector.transform(X_test), columns=X_test.columns[selector.get_support()], index=X_test.index)
    selected_features = X_train.columns.tolist()

    model_defs = [('Linear Regression', LinearRegression()), ('Ridge', Ridge(random_state=42)), ('Lasso', Lasso(random_state=42)), ('Random Forest', RandomForestRegressor(random_state=42, n_jobs=-1)), ('Gradient Boosting', GradientBoostingRegressor(random_state=42))]
    if _has_extra_libs: model_defs.extend([('XGBoost', xgb.XGBRegressor(random_state=42, n_jobs=-1, verbosity=0)), ('LightGBM', lgb.LGBMRegressor(random_state=42, n_jobs=-1, verbosity=-1)), ('CatBoost', cb.CatBoostRegressor(random_state=42, verbose=0))])
    
    results_list, trained_models = [], {}
    for i, (name, model) in enumerate(model_defs):
        progress(0.2 + (i / len(model_defs)) * 0.8, desc=f"Training {name}...")
        yield f"Training {i+1}/{len(model_defs)}: {name}...", None, None
        start_time = time.time(); model.fit(X_train, y_train); y_pred = model.predict(X_test)
        results_list.append({'Model': name, 'R²': r2_score(y_test, y_pred), 'MAE': mean_absolute_error(y_test, y_pred), 'RMSE': np.sqrt(mean_squared_error(y_test, y_pred)), 'Time (s)': f"{time.time() - start_time:.2f}"})
        trained_models[name] = model

    results_df = pd.DataFrame(results_list).sort_values(by='R²', ascending=False).reset_index(drop=True)
    plotter = ModelPlotter(trained_models, X_test, y_test)
    model_run_results = ModelRunResult(results_df, plotter, trained_models, selected_features)
    
    model_choices = results_df['Model'].tolist()
    yield "✅ Model training & evaluation complete.", model_run_results, gr.Dropdown(choices=model_choices, interactive=True)

# Replacement for mols2grid.display in Step 3
def create_prediction_grid_html(df, smiles_col='canonical_smiles', pred_col='predicted_pIC50', max_mols=20):
    html_parts = ['<div style="display: flex; flex-wrap: wrap; gap: 10px;">']
    for idx, row in df.head(max_mols).iterrows():
        smiles = row[smiles_col]
        pred_pic50 = row[pred_col]
        if pd.isna(pred_pic50):
            continue
        mol = Chem.MolFromSmiles(smiles)
        if mol:
            # Generate molecule image
            img = Draw.MolToImage(mol, size=(200, 200))
            # Convert to base64
            buffered = BytesIO()
            img.save(buffered, format="PNG")
            img_str = base64.b64encode(buffered.getvalue()).decode()
            # Create HTML for this molecule
            mol_html = f'''
            <div style="border: 1px solid #ccc; padding: 10px; border-radius: 5px; text-align: center;">
                <img src="data:image/png;base64,{img_str}" alt="Molecule" style="max-width: 200px;">
                <p><strong>Predicted pIC50:</strong> {pred_pic50:.2f}</p>
                <p style="font-size: 10px; word-break: break-all;">{smiles}</p>
            </div>
            '''
            html_parts.append(mol_html)
    html_parts.append('</div>')
    return ''.join(html_parts)

def predict_on_upload(uploaded_file, model_name, current_state, progress=gr.Progress()):
    if not uploaded_file: raise gr.Error("Please upload a file.")
    if not model_name: raise gr.Error("Please select a trained model.")
    model_run_results = current_state.get('model_results')
    fingerprint_type = current_state.get('fingerprint_type')
    if not model_run_results or not fingerprint_type: raise gr.Error("Please run Steps 2 and 3 first.")
    
    model = model_run_results.models.get(model_name)
    selected_features = model_run_results.selected_features
    if model is None: raise gr.Error(f"Model '{model_name}' not found.")
    
    smi_file, output_csv = 'predict.smi', 'predict_fp.csv'
    try:
        progress(0, desc="Reading & processing new molecules..."); yield "Reading uploaded file...", None, None
        df_new = pd.read_csv(uploaded_file.name)
        if 'canonical_smiles' not in df_new.columns: raise gr.Error("CSV must contain a 'canonical_smiles' column.")
        df_new = df_new.reset_index().rename(columns={'index': 'mol_id'})
        
        padel_input = pd.DataFrame({'smiles': df_new['canonical_smiles'], 'name': df_new['mol_id']})
        padel_input.to_csv(smi_file, sep='\t', index=False, header=False)
        if os.path.exists(output_csv): os.remove(output_csv)
        
        progress(0.3, desc="Calculating fingerprints..."); yield "Calculating fingerprints for new molecules...", None, None
        padeldescriptor(mol_dir=smi_file, d_file=output_csv, descriptortypes=fp_config.get(fingerprint_type), detectaromaticity=True, standardizenitro=True, threads=-1, removesalt=True, log=False, fingerprints=True)
        if not os.path.exists(output_csv) or os.path.getsize(output_csv) == 0: raise gr.Error("PaDEL calculation failed for the uploaded molecules.")
        
        progress(0.7, desc="Aligning features and predicting..."); yield "Aligning features and predicting...", None, None
        df_fp = pd.read_csv(output_csv).rename(columns={'Name': 'mol_id'})
        
        X_new = df_fp.set_index('mol_id')
        X_new_aligned = X_new.reindex(columns=selected_features, fill_value=0)[selected_features]
        
        predictions = model.predict(X_new_aligned)
        
        results_subset = pd.DataFrame({'mol_id': X_new_aligned.index, 'predicted_pIC50': predictions})
        df_results = pd.merge(df_new, results_subset, on='mol_id', how='left')

        progress(0.9, desc="Generating visualization..."); yield "Generating visualization...", None, None
        
        # Replacement for mols22grid.display in Step 3
        df_grid_view = df_results.dropna(subset=['predicted_pIC50']).copy()
        mols_html = "<h3>No molecules with successful predictions to display.</h3>"
        if not df_grid_view.empty:
            mols_html = create_prediction_grid_html(df_grid_view)
        
        progress(1, desc="Complete!"); yield "✅ Prediction complete.", df_results[['canonical_smiles', 'predicted_pIC50']], mols_html
    finally:
        if os.path.exists(smi_file): os.remove(smi_file)
        if os.path.exists(output_csv): os.remove(output_csv)

# ==============================================================================
# === GRADIO INTERFACE ===
# ==============================================================================
with gr.Blocks(theme=gr.themes.Default(primary_hue="blue", secondary_hue="sky"), title="Comprehensive Drug Discovery Workflow") as demo:
    gr.Markdown("# 🧪 Comprehensive Drug Discovery Workflow")
    gr.Markdown("A 3-step application to fetch, analyze, and model chemical bioactivity data.")
    app_state = gr.State({})
    with gr.Tabs():
        with gr.Tab("Step 1: Data Collection & EDA"):
            gr.Markdown("## Fetch Bioactivity Data from ChEMBL and Perform Exploratory Analysis")
            gr.Markdown(
                "This app allows you to fetch bioactivity data, perform exploratory data analysis, "
                "engineer molecular features, and train machine learning models for drug discovery. "
                "For an efficient example, use 'coronavirus' as the target query and select 'CHEMBL3927'."
            )
            with gr.Row():
                query_input = gr.Textbox(label="Target Query", placeholder="e.g., acetylcholinesterase, BRAF kinase", scale=3)
                fetch_btn = gr.Button("Fetch Targets", variant="primary", scale=1)
            status_step1_fetch = gr.Textbox(label="Status", interactive=False)
            target_id_table = gr.Dataframe(label="Available Targets", interactive=False, headers=["target_chembl_id", "pref_name", "organism"])
            with gr.Row():
                selected_target_dropdown = gr.Dropdown(label="Select Target ChEMBL ID", interactive=True, scale=3)
                process_btn = gr.Button("Process Data & Run EDA", variant="primary", scale=1, interactive=False)
            status_step1_process = gr.Textbox(label="Status", interactive=False)
            gr.Markdown("### Filtered Data & Analysis")
            bioactivity_class_selector = gr.CheckboxGroup(["active", "inactive", "intermediate"], label="Filter by Bioactivity Class", value=["active", "inactive", "intermediate"])
            df_output_s1 = gr.Dataframe(label="Cleaned Bioactivity Data")
            with gr.Tabs():
                with gr.Tab("Chemical Space Overview"):
                    with gr.Row():
                        freq_plot_output = gr.Plot(label="Frequency of Bioactivity Classes")
                        scatter_plot_output = gr.Plot(label="Scatter Plot: MW vs LogP")
                with gr.Tab("pIC50 Analysis"):
                    with gr.Row():
                        pic50_plot_output = gr.Plot(label="pIC50 Box Plot")
                        pic50_stats_output = gr.Dataframe(label="Mann-Whitney U Test Results for pIC50")
                with gr.Tab("Molecular Weight Analysis"):
                    with gr.Row():
                        mw_plot_output = gr.Plot(label="MW Box Plot")
                        mw_stats_output = gr.Dataframe(label="Mann-Whitney U Test Results for MW")
                with gr.Tab("LogP Analysis"):
                    with gr.Row():
                        logp_plot_output = gr.Plot(label="LogP Box Plot")
                        logp_stats_output = gr.Dataframe(label="Mann-Whitney U Test Results for LogP")
                with gr.Tab("H-Bond Donor/Acceptor Analysis"):
                    with gr.Row():
                        hdonors_plot_output = gr.Plot(label="H-Donors Box Plot")
                        hacceptors_plot_output = gr.Plot(label="H-Acceptors Box Plot")
                    with gr.Row():
                        hdonors_stats_output = gr.Dataframe(label="Stats for H-Donors")
                        hacceptors_stats_output = gr.Dataframe(label="Stats for H-Acceptors")
        with gr.Tab("Step 2: Feature Engineering"):
            # UI Definition for Step 2...
            gr.Markdown("## Calculate Molecular Fingerprints using PaDEL")
            with gr.Row():
                fingerprint_dropdown = gr.Dropdown(choices=FP_list, value='PubChem' if 'PubChem' in FP_list else None, label="Select Fingerprint Method", scale=3)
                calculate_fp_btn = gr.Button("Calculate Fingerprints", variant="primary", scale=1)
            status_step2 = gr.Textbox(label="Status", interactive=False)
            output_df_s2 = gr.Dataframe(label="Final Processed Data", wrap=True)
            download_s2 = gr.DownloadButton("Download Feature Data (CSV)", variant="secondary", visible=False)
            mols_grid_s2 = gr.HTML(label="Interactive Molecule Viewer")
        with gr.Tab("Step 3: Model Training & Prediction"):
            # UI Definition for Step 3...
            gr.Markdown("## Train Regression Models and Predict pIC50")
            with gr.Tabs():
                with gr.Tab("Model Training & Evaluation"):
                    train_models_btn = gr.Button("Train All Models", variant="primary")
                    status_step3_train = gr.Textbox(label="Status", interactive=False)
                    model_results_df = gr.DataFrame(label="Ranked Model Results", interactive=False)
                    with gr.Row():
                        model_selector_s3 = gr.Dropdown(label="Select Model to Analyze", interactive=False)
                        feature_count_s3 = gr.Number(label="Top Features to Show", value=7, minimum=3, maximum=20, step=1)
                    with gr.Tabs():
                        with gr.Tab("Validation Plots"): validation_plot_s3 = gr.Plot(label="Model Validation Plots")
                        with gr.Tab("Feature Importance"): feature_plot_s3 = gr.Plot(label="Top Feature Importances")
                with gr.Tab("Predict on New Data"):
                    gr.Markdown("Upload a CSV with a `canonical_smiles` column to predict pIC50.")
                    with gr.Row():
                        upload_predict_file = gr.File(label="Upload CSV for Prediction", file_types=[".csv"])
                        predict_btn_s3 = gr.Button("Run Prediction", variant="primary")
                    status_step3_predict = gr.Textbox(label="Status", interactive=False)
                    prediction_results_df = gr.DataFrame(label="Prediction Results")
                    prediction_mols_grid = gr.HTML(label="Interactive Molecular Grid of Predictions")

    # --- EVENT HANDLERS ---
    def enable_process_button(target_id): return gr.update(interactive=bool(target_id))
    def process_and_analyze_wrapper(target_id, selected_classes, current_state, progress=gr.Progress()):
        if not target_id: raise gr.Error("Please select a target ChEMBL ID first.")
        progress(0, desc="Fetching data..."); raw_data, msg1 = get_bioactivity_data(target_id); yield {status_step1_process: gr.update(value=msg1)}
        progress(0.3, desc="Cleaning data..."); processed_data, msg2 = clean_and_process_data(raw_data); yield {df_output_s1: processed_data, status_step1_process: gr.update(value=msg2)}
        current_state['cleaned_data'] = processed_data
        progress(0.6, desc="Running EDA..."); plots_and_stats = run_eda_analysis(processed_data, selected_classes); msg3 = plots_and_stats[-1]
        progress(1, desc="Done!")
        filtered_data = processed_data[processed_data.bioactivity_class.isin(selected_classes)] if not processed_data.empty else pd.DataFrame()
        outputs = [filtered_data] + list(plots_and_stats[:-1]) + [msg3, current_state]
        output_components = [df_output_s1, freq_plot_output, scatter_plot_output, pic50_plot_output, pic50_stats_output, mw_plot_output, mw_stats_output, logp_plot_output, logp_stats_output, hdonors_plot_output, hdonors_stats_output, hacceptors_plot_output, hacceptors_stats_output, status_step1_process, app_state]
        yield dict(zip(output_components, outputs))
    def update_analysis_on_filter_change(selected_classes, current_state):
        cleaned_data = current_state.get('cleaned_data')
        if cleaned_data is None or cleaned_data.empty: return (pd.DataFrame(),) + (None,) * 11 + ("No data available.",)
        plots_and_stats = run_eda_analysis(cleaned_data, selected_classes); msg = plots_and_stats[-1]
        filtered_data = cleaned_data[cleaned_data.bioactivity_class.isin(selected_classes)]
        return (filtered_data,) + plots_and_stats[:-1] + (msg,)
    def handle_model_training(current_state, progress=gr.Progress(track_tqdm=True)):
        fingerprint_data = current_state.get('fingerprint_data')
        if fingerprint_data is None or fingerprint_data.empty: raise gr.Error("No feature data. Please complete Step 2.")
        for status_msg, model_results, model_choices_update in run_regression_suite(fingerprint_data, progress=progress):
            if model_results: current_state['model_results'] = model_results
            yield status_msg, model_results.dataframe if model_results else None, model_choices_update, current_state
    def save_dataframe_as_csv(df):
        if df is None or df.empty: return None
        filename = "feature_engineered_data.csv"; df.to_csv(filename, index=False); return gr.File(value=filename, visible=True)
    def update_analysis_plots(model_name, feature_count, current_state):
        model_results = current_state.get('model_results')
        if not model_results or not model_name: return None, None
        plotter = model_results.plotter; validation_fig = plotter.plot_validation(model_name); feature_fig = plotter.plot_feature_importance(model_name, int(feature_count)); plt.close('all'); return validation_fig, feature_fig

    fetch_btn.click(fn=get_target_chembl_id, inputs=query_input, outputs=[target_id_table, selected_target_dropdown, status_step1_fetch], show_progress="minimal")
    selected_target_dropdown.change(fn=enable_process_button, inputs=selected_target_dropdown, outputs=process_btn, show_progress="hidden")
    process_btn.click(fn=process_and_analyze_wrapper, inputs=[selected_target_dropdown, bioactivity_class_selector, app_state], outputs=[df_output_s1, freq_plot_output, scatter_plot_output, pic50_plot_output, pic50_stats_output, mw_plot_output, mw_stats_output, logp_plot_output, logp_stats_output, hdonors_plot_output, hdonors_stats_output, hacceptors_plot_output, hacceptors_stats_output, status_step1_process, app_state])
    bioactivity_class_selector.change(fn=update_analysis_on_filter_change, inputs=[bioactivity_class_selector, app_state], outputs=[df_output_s1, freq_plot_output, scatter_plot_output, pic50_plot_output, pic50_stats_output, mw_plot_output, mw_stats_output, logp_plot_output, logp_stats_output, hdonors_plot_output, hdonors_stats_output, hacceptors_plot_output, hacceptors_stats_output, status_step1_process], show_progress="minimal")
    calculate_fp_btn.click(fn=calculate_fingerprints, inputs=[app_state, fingerprint_dropdown], outputs=[status_step2, output_df_s2, download_s2, mols_grid_s2, app_state])
    # The download button click handler was incorrect, it should take the dataframe from the state
    @download_s2.click(inputs=app_state, outputs=download_s2, show_progress="hidden")
    def download_handler(current_state):
        df_to_download = current_state.get('fingerprint_data')
        return save_dataframe_as_csv(df_to_download)
    train_models_btn.click(fn=handle_model_training, inputs=[app_state], outputs=[status_step3_train, model_results_df, model_selector_s3, app_state])
    for listener in [model_selector_s3.change, feature_count_s3.change]: listener(fn=update_analysis_plots, inputs=[model_selector_s3, feature_count_s3, app_state], outputs=[validation_plot_s3, feature_plot_s3], show_progress="minimal")
    predict_btn_s3.click(fn=predict_on_upload, inputs=[upload_predict_file, model_selector_s3, app_state], outputs=[status_step3_predict, prediction_results_df, prediction_mols_grid])

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