Update app.py

app.py

@@ -1,70 +1,69 @@
 # --- 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 #
+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 #
-import mols2grid #
+from rdkit import Chem
+from rdkit.Chem import Descriptors, Lipinski
+from chembl_webresource_client.new_client import new_client
+from padelpy import padeldescriptor
+# import mols2grid # This line will be removed
 
 # Plotting and Visualization
-import matplotlib.pyplot as plt #
-import seaborn as sns #
-from scipy import stats #
-from scipy.stats import mannwhitneyu #
+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.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.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 #
+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 #
+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.") #
+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') #
+warnings.filterwarnings("ignore")
+sns.set_theme(style='whitegrid')
 
 # --- FINGERPRINT CONFIGURATION ---
 DESCRIPTOR_DIR = "padel_descriptors"
-
 # Check if the descriptor directory exists and contains files
 if not os.path.isdir(DESCRIPTOR_DIR):
     warnings.warn(
@@ -74,411 +73,893 @@ if not os.path.isdir(DESCRIPTOR_DIR):
     xml_files = []
 else:
     xml_files = sorted(glob.glob(os.path.join(DESCRIPTOR_DIR, '*.xml')))
-
-if not xml_files:
-    warnings.warn(
-        f"No descriptor .xml files found in the '{DESCRIPTOR_DIR}' directory. "
-        "Fingerprint calculation will not be possible."
-    )
+    if not xml_files:
+        warnings.warn(
+            f"No descriptor .xml files found in the '{DESCRIPTOR_DIR}' directory. "
+            "Fingerprint calculation will not be possible."
+        )
 
 # The key is the filename without extension; the value is the full path to the file
 fp_config = {os.path.splitext(os.path.basename(file))[0]: file for file in xml_files}
 FP_list = sorted(list(fp_config.keys()))
 
 
+# --- NEW MOLECULE DISPLAY FUNCTIONS ---
+def create_molecule_grid_html(df, smiles_col='canonical_smiles', additional_cols=None, max_mols=50):
+    """
+    Create a custom HTML grid for displaying molecules using RDKit and base64 encoding
+    This works better in Hugging Face Spaces than mols2grid
+    """
+    from rdkit import Chem
+    from rdkit.Chem import Draw
+    import base64
+    from io import BytesIO
+    
+    if df.empty:
+        return "<h3>No molecules to display.</h3>"
+    
+    # Limit number of molecules for performance
+    df_display = df.head(max_mols).copy()
+    
+    # Additional columns to display
+    if additional_cols is None:
+        additional_cols = []
+    
+    html_parts = ["""
+    <div style="display: grid; grid-template-columns: repeat(auto-fill, minmax(300px, 1fr)); gap: 15px; padding: 10px;">
+    """]
+    
+    for idx, row in df_display.iterrows():
+        try:
+            # Generate molecule image
+            mol = Chem.MolFromSmiles(row[smiles_col])
+            if mol is None:
+                continue
+                
+            # Draw molecule
+            img = Draw.MolToImage(mol, size=(250, 200))
+            
+            # Convert to base64
+            buffer = BytesIO()
+            img.save(buffer, format='PNG')
+            img_str = base64.b64encode(buffer.getvalue()).decode()
+            
+            # Create card HTML
+            card_html = f"""
+            <div style="border: 1px solid #ddd; border-radius: 8px; padding: 10px; background: white;">
+                <img src="data:image/png;base64,{img_str}" style="width: 100%; height: auto;" alt="Molecule"/>
+                <div style="margin-top: 8px; font-size: 12px;">
+                    <strong>SMILES:</strong> {row[smiles_col][:50]}{'...' if len(str(row[smiles_col])) > 50 else ''}<br/>
+            """
+            
+            # Add additional columns
+            for col in additional_cols:
+                if col in row and pd.notna(row[col]):
+                    value = row[col]
+                    if isinstance(value, float):
+                        value = f"{value:.2f}"
+                    card_html += f"<strong>{col}:</strong> {value}<br/>"
+            
+            card_html += """
+                </div>
+            </div>
+            """
+            html_parts.append(card_html)
+            
+        except Exception as e:
+            print(f"Error processing molecule {idx}: {e}")
+            continue
+    
+    html_parts.append("</div>")
+    
+    if len(html_parts) == 2:  # Only header and footer, no molecules processed
+        return "<h3>No valid molecules to display.</h3>"
+    
+    return "".join(html_parts)
+
+def create_simple_molecule_table(df, smiles_col='canonical_smiles', additional_cols=None, max_mols=20):
+    """
+    Create a simple HTML table with molecule images - fallback option
+    """
+    from rdkit import Chem
+    from rdkit.Chem import Draw
+    import base64
+    from io import BytesIO
+    
+    if df.empty:
+        return "<h3>No molecules to display.</h3>"
+    
+    df_display = df.head(max_mols).copy()
+    
+    if additional_cols is None:
+        additional_cols = []
+    
+    # Start HTML table
+    html = """
+    <table style="border-collapse: collapse; width: 100%;">
+        <thead>
+            <tr style="background-color: #f2f2f2;">
+                <th style="border: 1px solid #ddd; padding: 8px;">Structure</th>
+                <th style="border: 1px solid #ddd; padding: 8px;">SMILES</th>
+    """
+    
+    for col in additional_cols:
+        html += f'<th style="border: 1px solid #ddd; padding: 8px;">{col}</th>'
+    
+    html += """
+            </tr>
+        </thead>
+        <tbody>
+    """
+    
+    for idx, row in df_display.iterrows():
+        try:
+            mol = Chem.MolFromSmiles(row[smiles_col])
+            if mol is None:
+                continue
+                
+            # Generate image
+            img = Draw.MolToImage(mol, size=(200, 150))
+            buffer = BytesIO()
+            img.save(buffer, format='PNG')
+            img_str = base64.b64encode(buffer.getvalue()).decode()
+            
+            html += f"""
+            <tr>
+                <td style="border: 1px solid #ddd; padding: 8px; text-align: center;">
+                    <img src="data:image/png;base64,{img_str}" style="max-width: 200px; height: auto;" alt="Molecule"/>
+                </td>
+                <td style="border: 1px solid #ddd; padding: 8px; font-family: monospace; font-size: 11px;">
+                    {row[smiles_col][:100]}{'...' if len(str(row[smiles_col])) > 100 else ''}
+                </td>
+            """
+            
+            for col in additional_cols:
+                value = row[col] if col in row and pd.notna(row[col]) else "N/A"
+                if isinstance(value, float):
+                    value = f"{value:.2f}"
+                html += f'<td style="border: 1px solid #ddd; padding: 8px;">{value}</td>'
+            
+            html += "</tr>"
+            
+        except Exception as e:
+            print(f"Error processing molecule {idx}: {e}")
+            continue
+    
+    html += "</tbody></table>"
+    return html
+
 # ==============================================================================
 # === 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") #
+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()] #
-    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) #
+    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()]
+    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 ===
 # ==============================================================================
-
-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' #
+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'
+        
+        input_df[['canonical_smiles', 'canonical_smiles']].to_csv(smi_file, sep='\t', index=False, header=False)
         
-        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.")
         
-        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
         
-        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.") #
+        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'}) #
+        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'})
+        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...")
+        
+        # Use custom molecule display instead of mols2grid
+        try:
+            # Try the grid layout first
+            mols_html = create_molecule_grid_html(
+                final_df,  
+                smiles_col='canonical_smiles',  
+                additional_cols=['pIC50'],
+                max_mols=50
+            )
+        except Exception as e:
+            print(f"Grid layout failed: {e}, trying table layout...")
+            # Fallback to table layout
+            mols_html = create_simple_molecule_table(
+                final_df,
+                smiles_col='canonical_smiles',
+                additional_cols=['pIC50'],
+                max_mols=20
+            )
         
-        final_df = pd.merge(input_df[['canonical_smiles', 'pIC50']], df_X, on='canonical_smiles', how='inner') #
+        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
         
-        current_state['fingerprint_data'] = final_df; current_state['fingerprint_type'] = fingerprint_type #
-        progress(0.9, desc="🖼️ Generating molecule grid...") #
-        mols_html = mols2grid.display(final_df, smiles_col='canonical_smiles', subset=['img', 'pIC50'], rename={"pIC50": "pIC50"}, transform={"pIC50": lambda x: f"{x:.2f}"})._repr_html_() #
-        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') #
+    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 ===
+# === STEP 3: MODELING 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)) # <-- Commented out
-    ]
-    if _has_extra_libs:
-        model_defs.extend([
-            # ('XGBoost', xgb.XGBRegressor(random_state=42, n_jobs=-1, verbosity=0)), # <-- Commented out
-            ('LightGBM', lgb.LGBMRegressor(random_state=42, n_jobs=-1, verbosity=-1)),
-            # ('CatBoost', cb.CatBoostRegressor(random_state=42, verbose=0)) # <-- Commented out
-        ])
+
+# Model definitions
+regression_models = {
+    "Linear Regression": LinearRegression,
+    "Ridge Regression": Ridge,
+    "Lasso Regression": Lasso,
+    "Elastic Net": ElasticNet,
+    "Bayesian Ridge": BayesianRidge,
+    "Huber Regressor": HuberRegressor,
+    "Passive Aggressive Regressor": PassiveAggressiveRegressor,
+    "Orthogonal Matching Pursuit": OrthogonalMatchingPursuit,
+    "Lasso Lars": LassoLars,
+    "Decision Tree Regressor": DecisionTreeRegressor,
+    "Random Forest Regressor": RandomForestRegressor,
+    "Gradient Boosting Regressor": GradientBoostingRegressor,
+    "Extra Trees Regressor": ExtraTreesRegressor,
+    "AdaBoost Regressor": AdaBoostRegressor,
+    "K-Neighbors Regressor": KNeighborsRegressor,
+    "Dummy Regressor (Mean)": DummyRegressor,
+}
+
+if _has_extra_libs:
+    regression_models.update({
+        "XGBoost Regressor": xgb.XGBRegressor,
+        "LightGBM Regressor": lgb.LGBMRegressor,
+        "CatBoost Regressor": cb.CatBoostRegressor,
+    })
+
+def handle_model_training(current_state, progress=gr.Progress()):
+    df_fp = current_state.get('fingerprint_data')
+    if df_fp is None or df_fp.empty:
+        raise gr.Error("No fingerprint data found. Please complete Step 2.")
     
-    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) #
+    yield "⚙️ Starting model training...", None, None, current_state
+    progress(0, desc="Starting model training...")
+
+    try:
+        X = df_fp.drop(columns=['canonical_smiles', 'pIC50'])
+        y = df_fp['pIC50']
+
+        if X.empty:
+            raise gr.Error("No features found for training. Check fingerprint calculation.")
+
+        # Remove features with zero variance
+        sel = VarianceThreshold(threshold=0.0)
+        X_filtered = sel.fit_transform(X)
+        selected_features = X.columns[sel.get_support()]
+
+        if len(selected_features) == 0:
+            raise gr.Error("No features with non-zero variance found. Cannot train models.")
+
+        X_filtered = pd.DataFrame(X_filtered, columns=selected_features)
+
+        X_train, X_test, y_train, y_test = train_test_split(X_filtered, y, test_size=0.2, random_state=42)
+
+        results = []
+        trained_models = {}
+
+        for i, (name, model_class) in enumerate(regression_models.items()):
+            current_progress = (i + 1) / len(regression_models)
+            progress(current_progress, desc=f"Training {name}...")
+            yield f"Training {name}...", None, None, current_state
+            
+            model = model_class()
+            model.fit(X_train, y_train)
+            y_pred = model.predict(X_test)
+            
+            mae = mean_absolute_error(y_test, y_pred)
+            mse = mean_squared_error(y_test, y_pred)
+            r2 = r2_score(y_test, y_pred)
+            
+            results.append({
+                "Model": name,
+                "MAE": mae,
+                "MSE": mse,
+                "R2": r2
+            })
+            trained_models[name] = model
+
+        results_df = pd.DataFrame(results)
+        
+        # Store results in the state
+        current_state['model_results'] = ModelRunResult(
+            dataframe=results_df,
+            plotter=None, # No plot generated here directly, but could be added
+            models=trained_models,
+            selected_features=selected_features
+        )
+        current_state['X_train'] = X_train
+        current_state['y_train'] = y_train
+        current_state['X_test'] = X_test
+        current_state['y_test'] = y_test
+
+        progress(1, desc="Model training complete!")
+        yield "✅ Model training complete!", results_df, gr.update(choices=list(trained_models.keys()), value=list(trained_models.keys())[0]), current_state
+
+    except Exception as e:
+        raise gr.Error(f"Model training failed: {e}")
+
+def update_analysis_plots(model_name, feature_count, current_state):
+    model_run_results = current_state.get('model_results')
+    X_train = current_state.get('X_train')
+    y_train = current_state.get('y_train')
+    X_test = current_state.get('X_test')
+    y_test = current_state.get('y_test')
+
+    if not model_run_results or not model_name or X_test is None or y_test is None:
+        return None, None, None, None, "Please train models first."
+
+    model = model_run_results.models.get(model_name)
+    if model is None:
+        return None, None, None, None, f"Model '{model_name}' not found."
+
+    y_pred_test = model.predict(X_test)
+    y_pred_train = model.predict(X_train)
+
+    # Calculate R2, MAE, MSE for test set
+    r2_test = r2_score(y_test, y_pred_test)
+    mae_test = mean_absolute_error(y_test, y_pred_test)
+    mse_test = mean_squared_error(y_test, y_pred_test)
+
+    # Plot Y-obs vs Y-pred
+    fig_obs_pred, ax_obs_pred = plt.subplots(figsize=(6, 6))
+    sns.scatterplot(x=y_test, y=y_pred_test, ax=ax_obs_pred, alpha=0.7, color='blue', label='Test Data')
+    sns.scatterplot(x=y_train, y=y_pred_train, ax=ax_obs_pred, alpha=0.7, color='green', label='Train Data')
+    ax_obs_pred.plot([min(y_test.min(), y_train.min()), max(y_test.max(), y_train.max())],
+                     [min(y_test.min(), y_train.min()), max(y_test.max(), y_train.max())],
+                     color='red', linestyle='--', label='Ideal Prediction')
+    ax_obs_pred.set_xlabel("Observed pIC50", fontsize=12)
+    ax_obs_pred.set_ylabel("Predicted pIC50", fontsize=12)
+    ax_obs_pred.set_title(f"{model_name}: Observed vs. Predicted pIC50", fontsize=14)
+    ax_obs_pred.legend()
+    plt.close(fig_obs_pred)
+
+    # Plot Residuals
+    residuals = y_test - y_pred_test
+    fig_residuals, ax_residuals = plt.subplots(figsize=(6, 6))
+    sns.scatterplot(x=y_pred_test, y=residuals, ax=ax_residuals, alpha=0.7, color='purple')
+    ax_residuals.axhline(y=0, color='red', linestyle='--')
+    ax_residuals.set_xlabel("Predicted pIC50", fontsize=12)
+    ax_residuals.set_ylabel("Residuals (Observed - Predicted)", fontsize=12)
+    ax_residuals.set_title(f"{model_name}: Residuals Plot", fontsize=14)
+    plt.close(fig_residuals)
+
+    # Feature Importance Plot (if applicable)
+    fig_feature_importance = None
+    if hasattr(model, 'feature_importances_') and feature_count > 0:
+        feature_importances = pd.Series(model.feature_importances_, index=model_run_results.selected_features)
+        top_features = feature_importances.nlargest(feature_count)
+
+        fig_feature_importance, ax_fi = plt.subplots(figsize=(8, 6))
+        sns.barplot(x=top_features.values, y=top_features.index, ax=ax_fi, palette='viridis')
+        ax_fi.set_xlabel("Importance", fontsize=12)
+        ax_fi.set_ylabel("Feature", fontsize=12)
+        ax_fi.set_title(f"{model_name}: Top {feature_count} Feature Importances", fontsize=14)
+        plt.tight_layout()
+        plt.close(fig_feature_importance)
+    elif hasattr(model, 'coef_') and feature_count > 0 and model_name not in ["Dummy Regressor (Mean)", "K-Neighbors Regressor"]:
+        # For linear models, coefficients can be used as importance
+        feature_importances = pd.Series(model.coef_, index=model_run_results.selected_features)
+        top_features = feature_importances.abs().nlargest(feature_count) # Use absolute value for ranking
+        top_features_values = feature_importances[top_features.index] # Get actual signed values
+
+        fig_feature_importance, ax_fi = plt.subplots(figsize=(8, 6))
+        sns.barplot(x=top_features_values.values, y=top_features_values.index, ax=ax_fi, palette='coolwarm')
+        ax_fi.set_xlabel("Coefficient Value", fontsize=12)
+        ax_fi.set_ylabel("Feature", fontsize=12)
+        ax_fi.set_title(f"{model_name}: Top {feature_count} Feature Coefficients", fontsize=14)
+        plt.tight_layout()
+        plt.close(fig_feature_importance)
+
+
+    return fig_obs_pred, fig_residuals, fig_feature_importance, \
+           f"R2 (Test): {r2_test:.4f}, MAE (Test): {mae_test:.4f}, MSE (Test): {mse_test:.4f}", \
+           "Plots updated."
+
+# Updated prediction function
+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_choices = results_df['Model'].tolist() #
-    yield "✅ Model training & evaluation complete.", model_run_results, gr.Dropdown(choices=model_choices, interactive=True) #
-
-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_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.") #
+    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'}) #
+    smi_file, output_csv = 'predict.smi', 'predict_fp.csv'
+    try:
+        progress(0, desc="Reading & processing new molecules...")
+        yield "Reading uploaded file...", None, None
         
-        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) #
+        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'})
         
-        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.") #
+        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(0.7, desc="Aligning features and predicting..."); yield "Aligning features and predicting...", None, None #
-        df_fp = pd.read_csv(output_csv).rename(columns={'Name': 'mol_id'}) #
+        progress(0.3, desc="Calculating fingerprints...")
+        yield "Calculating fingerprints for new molecules...", None, None
         
-        X_new = df_fp.set_index('mol_id') #
-        X_new_aligned = X_new.reindex(columns=selected_features, fill_value=0)[selected_features] #
+        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
+        )
         
-        predictions = model.predict(X_new_aligned) #
+        if not os.path.exists(output_csv) or os.path.getsize(output_csv) == 0:  
+            raise gr.Error("PaDEL calculation failed for the uploaded molecules.")
         
-        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 #
+        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
         
-        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: #
-            df_grid_view.rename(columns={"predicted_pIC50": "Predicted pIC50"}, inplace=True) #
-            mols_html = mols2grid.display( #
-                df_grid_view, #
-                smiles_col='canonical_smiles', #
-                subset=['img', 'Predicted pIC50'], #
-                transform={"Predicted pIC50": lambda x: f"{x:.2f}"} #
-            )._repr_html_() #
+        df_grid_view = df_results.dropna(subset=['predicted_pIC50']).copy()
+        mols_html = "<h3>No molecules with successful predictions to display.</h3>"
         
-        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) #
+        if not df_grid_view.empty:
+            try:
+                # Use custom molecule display
+                mols_html = create_molecule_grid_html(
+                    df_grid_view,
+                    smiles_col='canonical_smiles',
+                    additional_cols=['predicted_pIC50'],
+                    max_mols=50
+                )
+            except Exception as e:
+                print(f"Grid layout failed: {e}, trying table layout...")
+                mols_html = create_simple_molecule_table(
+                    df_grid_view,
+                    smiles_col='canonical_smiles',
+                    additional_cols=['predicted_pIC50'],
+                    max_mols=20
+                )
+        
+        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 ===
+# === GRADIO INTERFACE LAYOUT ===
 # ==============================================================================
-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") #
-            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"): #
-            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"): #
-            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") #
+
+with gr.Blocks(css=".container { max-width: 1200px; margin: auto; }") as demo:
+    app_state = gr.State({}) # Store application state (e.g., fetched data, trained models)
+
+    gr.Markdown("# 💊 Bioactivity Prediction App")
+    gr.Markdown("---")
+
+    with gr.Tabs():
+        with gr.TabItem("Step 1: Data Collection & EDA"):
+            with gr.Row():
+                with gr.Column():
+                    gr.Markdown("## 1.1 ChEMBL Target Search")
+                    target_query = gr.Textbox(
+                        label="Enter target protein name (e.g., 'EGFR', 'acetylcholinesterase')",
+                        placeholder="EGFR"
+                    )
+                    search_target_btn = gr.Button("Search ChEMBL")
+                    target_output_df = gr.DataFrame(
+                        label="Search Results",
+                        headers=["target_chembl_id", "pref_name", "organism"],
+                        max_rows=5
+                    )
+                    status_step1_search = gr.Textbox(label="Status", interactive=False)
+
+                    gr.Markdown("## 1.2 Fetch Bioactivity Data")
+                    chembl_id_selector = gr.Dropdown(
+                        label="Select Target ChEMBL ID",
+                        choices=[], interactive=True
+                    )
+                    fetch_data_btn = gr.Button("Fetch Bioactivity Data (IC50)")
+                    bioactivity_output_df = gr.DataFrame(label="Raw Bioactivity Data", max_rows=5)
+                    status_step1_fetch = gr.Textbox(label="Status", interactive=False)
+
+                    gr.Markdown("## 1.3 Clean & Process Data")
+                    process_data_btn = gr.Button("Process & Calculate pIC50/Lipinski Descriptors")
+                    cleaned_data_output_df = gr.DataFrame(label="Cleaned & Processed Data", max_rows=5)
+                    status_step1_process_clean = gr.Textbox(label="Status", interactive=False)
+                    download_s1 = gr.DownloadButton("Download Cleaned Data (CSV)", visible=False, interactive=False)
+
+                with gr.Column():
+                    gr.Markdown("## 1.4 Exploratory Data Analysis (EDA)")
+                    bioactivity_class_selector = gr.CheckboxGroup(
+                        label="Select Bioactivity Classes for EDA",
+                        choices=["active", "inactive", "intermediate"],
+                        value=["active", "inactive", "intermediate"],
+                        interactive=True
+                    )
+                    run_eda_btn = gr.Button("Run EDA")
+                    status_step1_process = gr.Textbox(label="Status", interactive=False)
+
+                    gr.Markdown("### Bioactivity Class Frequency")
+                    freq_plot_output = gr.Plot(label="Bioactivity Class Frequency")
+
+                    gr.Markdown("### Chemical Space (MW vs LogP)")
+                    scatter_plot_output = gr.Plot(label="MW vs LogP Scatter Plot")
+
+                    gr.Markdown("### pIC50 Distribution")
+                    pic50_plot_output = gr.Plot(label="pIC50 Box Plot")
+                    pic50_stats_output = gr.DataFrame(label="pIC50 Mann-Whitney U Test", max_rows=5)
+
+                    gr.Markdown("### Molecular Weight (MW) Distribution")
+                    mw_plot_output = gr.Plot(label="MW Box Plot")
+                    mw_stats_output = gr.DataFrame(label="MW Mann-Whitney U Test", max_rows=5)
+
+                    gr.Markdown("### LogP Distribution")
+                    logp_plot_output = gr.Plot(label="LogP Box Plot")
+                    logp_stats_output = gr.DataFrame(label="LogP Mann-Whitney U Test", max_rows=5)
+
+                    gr.Markdown("### Hydrogen Donors Distribution")
+                    hdonors_plot_output = gr.Plot(label="Hydrogen Donors Box Plot")
+                    hdonors_stats_output = gr.DataFrame(label="Hydrogen Donors Mann-Whitney U Test", max_rows=5)
+
+                    gr.Markdown("### Hydrogen Acceptors Distribution")
+                    hacceptors_plot_output = gr.Plot(label="Hydrogen Acceptors Box Plot")
+                    hacceptors_stats_output = gr.DataFrame(label="Hydrogen Acceptors Mann-Whitney U Test", max_rows=5)
+
+        with gr.TabItem("Step 2: Feature Engineering (Fingerprints)"):
+            gr.Markdown("## 2.1 Calculate Molecular Fingerprints")
+            gr.Markdown(f"Available Fingerprint Types: {', '.join(FP_list)}")
+            fingerprint_dropdown = gr.Dropdown(
+                label="Select Fingerprint Type",
+                choices=FP_list,
+                interactive=True,
+                value=FP_list[0] if FP_list else None # Set default if available
+            )
+            calculate_fp_btn = gr.Button("Calculate Fingerprints")
+            status_step2 = gr.Textbox(label="Status", interactive=False)
+            output_df_s2 = gr.DataFrame(label="Fingerprint Data (First 5 rows, with pIC50)", max_rows=5)
+            download_s2 = gr.DownloadButton("Download Fingerprint Data (CSV)", visible=False, interactive=False)
+            
+            # Using HTML component for custom molecule display
+            mols_grid_s2 = gr.HTML(label="Molecules with pIC50", visible=True)
+
+
+        with gr.TabItem("Step 3: Model Training & Evaluation"):
+            gr.Markdown("## 3.1 Train Regression Models")
+            train_models_btn = gr.Button("Train Models")
+            status_step3_train = gr.Textbox(label="Status", interactive=False)
+            model_results_df = gr.DataFrame(label="Model Performance Metrics", max_rows=10)
+
+            gr.Markdown("## 3.2 Model Analysis")
+            model_selector_s3 = gr.Dropdown(
+                label="Select Model for Detailed Analysis",
+                choices=[],
+                interactive=True
+            )
+            feature_count_s3 = gr.Slider(
+                minimum=0, maximum=50, step=1, value=10,
+                label="Number of Top Features to Display", interactive=True
+            )
+            model_metrics_summary = gr.Textbox(label="Selected Model Metrics (Test Set)", interactive=False)
+            obs_pred_plot = gr.Plot(label="Observed vs. Predicted pIC50")
+            residuals_plot = gr.Plot(label="Residuals Plot")
+            feature_importance_plot = gr.Plot(label="Feature Importance/Coefficients")
+        
+        with gr.TabItem("Step 4: Prediction on New Data"):
+            gr.Markdown("## 4.1 Upload New Molecules & Predict")
+            upload_new_mols = gr.File(label="Upload CSV with 'canonical_smiles' column")
+            model_selector_s4 = gr.Dropdown(
+                label="Select Trained Model for Prediction",
+                choices=[],
+                interactive=True
+            )
+            predict_btn = gr.Button("Predict pIC50 for New Molecules")
+            status_step4 = gr.Textbox(label="Status", interactive=False)
+            predictions_output_df = gr.DataFrame(label="Predictions for New Molecules", max_rows=10)
+            download_s4 = gr.DownloadButton("Download Predictions (CSV)", visible=False, interactive=False)
+            # Using HTML component for custom molecule display
+            mols_grid_s4 = gr.HTML(label="New Molecules with Predicted pIC50", visible=True)
 
     # --- 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]) #
+
+    # Step 1 Callbacks
+    search_target_btn.click(
+        fn=lambda query: (get_target_chembl_id(query), gr.update(visible=True)), # Return two values
+        inputs=target_query,
+        outputs=[target_output_df, chembl_id_selector, status_step1_search]
+    )
+
+    chembl_id_selector.change(
+        fn=lambda x: gr.update(value=x),
+        inputs=chembl_id_selector,
+        outputs=chembl_id_selector # This is just to ensure the value is updated
+    )
+
+    fetch_data_btn.click(
+        fn=lambda target_id: (get_bioactivity_data(target_id), gr.update(value=target_id)),
+        inputs=chembl_id_selector,
+        outputs=[bioactivity_output_df, status_step1_fetch]
+    ).then(
+        fn=lambda df: (df, df.copy()), # Pass the fetched df to the state
+        inputs=bioactivity_output_df,
+        outputs=[gr.State(value={}, key='raw_data'), app_state]
+    )
+
+    process_data_btn.click(
+        fn=lambda current_state: clean_and_process_data(current_state.get('raw_data')),
+        inputs=app_state,
+        outputs=[cleaned_data_output_df, status_step1_process_clean]
+    ).then(
+        fn=lambda df: (gr.update(visible=True), df), # Show download button and update state
+        inputs=cleaned_data_output_df,
+        outputs=[download_s1, gr.State(value={}, key='cleaned_data'), app_state]
+    )
+
+    @download_s1.click(inputs=app_state, outputs=download_s1, show_progress="hidden")
+    def download_handler_s1(current_state):
+        df_to_download = current_state.get('cleaned_data')
+        if df_to_download is None:
+            raise gr.Error("No data to download. Please process data first.")
+        # Create a dummy file path for Gradio to handle the download
+        file_path = "cleaned_data.csv"
+        df_to_download.to_csv(file_path, index=False)
+        return gr.File(file_path, visible=True)
+
+    run_eda_btn.click(
+        fn=lambda df, classes: run_eda_analysis(df, classes),
+        inputs=[cleaned_data_output_df, bioactivity_class_selector],
+        outputs=[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]
+    )
     
-    @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) #
+    # Update EDA plots on filter change (if data is already processed)
+    bioactivity_class_selector.change(
+        fn=lambda current_state, selected_classes: run_eda_analysis(current_state.get('cleaned_data'), selected_classes),
+        inputs=[app_state, bioactivity_class_selector],
+        outputs=[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"
+    )
+
+
+    # Step 2 Callbacks
+    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]
+    )
     
-    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]) #
+    @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')
+        if df_to_download is None:
+            raise gr.Error("No data to download. Please calculate fingerprints first.")
+        file_path = "fingerprint_data.csv"
+        df_to_download.to_csv(file_path, index=False)
+        return gr.File(file_path, visible=True)
+
+
+    # Step 3 Callbacks
+    train_models_btn.click(
+        fn=handle_model_training,
+        inputs=[app_state],
+        outputs=[status_step3_train, model_results_df, model_selector_s3, app_state]
+    )
+
+    # Update plots when model or feature count changes
+    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=[obs_pred_plot, residuals_plot, feature_importance_plot, model_metrics_summary, status_step3_train]
+        )
+    
+    # Update model selector in Step 4 when models are trained
+    model_selector_s3.change(
+        fn=lambda choice: gr.update(choices=model_selector_s3.choices, value=choice),
+        inputs=model_selector_s3,
+        outputs=model_selector_s4
+    )
+
+
+    # Step 4 Callbacks
+    predict_btn.click(
+        fn=predict_on_upload,
+        inputs=[upload_new_mols, model_selector_s4, app_state],
+        outputs=[status_step4, predictions_output_df, mols_grid_s4]
+    ).then(
+        fn=lambda: gr.update(visible=True),
+        outputs=download_s4
+    )
+    
+    @download_s4.click(inputs=predictions_output_df, outputs=download_s4, show_progress="hidden")
+    def download_predictions(df_predictions):
+        if df_predictions is None or df_predictions.empty:
+            raise gr.Error("No predictions to download.")
+        file_path = "predictions.csv"
+        df_predictions.to_csv(file_path, index=False)
+        return gr.File(file_path, visible=True)
+
 
-if __name__ == "__main__": #
-    demo.launch(debug=True) #
+demo.launch()