app.py

import gradio as gr
from rdkit import Chem
from rdkit.Chem import Draw, Descriptors, AllChem
from rdkit.DataStructs.cDataStructs import ConvertToNumpyArray
import numpy as np
import xgboost as xgb
from PIL import Image


# Load the XGBoost model
model = xgb.XGBClassifier()
model.load_model("xg3.json")

# Function to convert SMILES string to descriptors and Lipinski features
def predict_smiles(smiles):
    mol = Chem.MolFromSmiles(smiles)
    if mol is None:  # Check if the conversion was unsuccessful
        return None, "Invalid SMILES string. Please enter a valid SMILES."

    fp = AllChem.GetMorganFingerprintAsBitVect(mol, radius=2, nBits=2048)
    fp_array = np.zeros((1,), dtype=int)
    ConvertToNumpyArray(fp, fp_array)
    lipinski_features = np.array([[Descriptors.MolWt(mol), Descriptors.MolLogP(mol), Descriptors.NumHAcceptors(mol), Descriptors.NumHDonors(mol)]])
    
    # Combine fingerprint and Lipinski features
    features = np.concatenate([fp_array, lipinski_features.flatten()])
    return features, lipinski_features

# Updated function to predict from SMILES and visualize Lipinski's features
def predict_and_visualize(smiles):
    features, lipinski_features = predict_smiles(smiles)
    if features is None:
        # Return a clear message for invalid SMILES, a placeholder for the features, and None for the image
        error_message = "Invalid SMILES string. Please enter a valid SMILES."
        placeholder_features = "<b><u>RDKit estimates</u></b><br>Not applicable due to invalid SMILES input."
        return error_message, placeholder_features, None
    
    # Process valid SMILES 
    molecular_weight, alogp, hba, hbd = lipinski_features.flatten()
    molecular_weight = round(molecular_weight, 1)
    alogp = round(alogp, 1)
    hba = int(hba)
    hbd = int(hbd)

    lipinski_features = np.array([[molecular_weight, alogp, hba, hbd]])

    prediction = model.predict(features.reshape(1, -1))
    result = "drug-like" if prediction == 1 else "not drug-like"
    
    mol = Chem.MolFromSmiles(smiles)
    img = Draw.MolToImage(mol)
    img = img.resize((800, 800), Image.Resampling.LANCZOS)  

    features_names = ["Molecular Weight", "AlogP", "HBA", "HBD"]
    lipinski_str = "<b><u>RDKit estimates</u></b><br>" + "<br>".join([f"{name}: {value}" for name, value in zip(features_names, [molecular_weight, alogp, hba, hbd])])
    
    return result, lipinski_str, img

# Gradio interface
iface = gr.Interface(fn=predict_and_visualize,
                     inputs=gr.Textbox(lines=2, placeholder="Enter SMILES string here..."),
                     outputs=[gr.Text(label="Prediction"),
                              gr.HTML(label="Lipinski's Features"),
                              gr.Image(label="Molecule Visualization")],
                     title="XGBoost Drug-like Classifier",
                     description="This application predicts whether a molecule is drug-like based on its SMILES representation.")

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