Bioactivity Prediction of Chemical Compounds Against Cyclooxygenase-2(COX-2) Using Machine Learning with Python

Overview

This project focuses on predicting the bioactivity of chemical compounds against Cyclooxygenase-2 (COX-2), an enzyme implicated in inflammation and various diseases. The goal is to develop a machine learning model that can accurately classify compounds as active or inactive COX-2 inhibitors, streamlining the early stages of drug discovery. The project leverages cheminformatics tools, Python-based machine learning, and a user-friendly web interface for predictions.

Problem Statement

Traditional drug discovery methods for COX-2 inhibitors are time-consuming, costly, and often yield compounds with undesirable side effects. This project addresses these challenges by:

• Reducing reliance on wet-lab experiments through computational predictions.
• Providing a user-friendly tool for researchers without extensive coding experience.
• Identifying potential COX-2 inhibitors with fewer side effects using machine learning.

Methodology

The project follows a structured pipeline from data collection to model deployment:

1. Data Collection and Preprocessing

• Data Source: Curated bioactivity data for COX-2 (Target ID: CHEMBL230) was extracted from the ChEMBL database.
• Preprocessing:
  • Removed rows with missing values.
  • Binarized activity labels: Compounds with IC50 < 10,000 nM were labeled as "active," others as "inactive."
  • Retained essential columns: molecule_chembl_id, SMILES, standard_value (IC50), and bioactivity_class.

2. Molecular Descriptor Calculation

• Tools: RDKit was used to compute six key molecular descriptors:
  • Molecular Weight (MolWt)
  • LogP (lipophilicity)
  • Hydrogen Bond Donors (HDonors)
  • Hydrogen Bond Acceptors (HAcceptors)
  • Topological Polar Surface Area (TPSA)
  • Number of Rotatable Bonds (NumRotatableBonds).
• Quality Control: Rows with invalid or missing descriptors were identified and excluded.

3. Machine Learning Model Training

• Algorithm: XGBoost (Extreme Gradient Boosting) was selected for its robustness in handling imbalanced datasets and non-linear relationships.
• Data Splitting: The dataset was split into 80% training and 20% testing sets, with stratification to maintain class distribution.
• Class Imbalance Handling: SMOTE-Tomek resampling was applied to balance the training data.
• Feature Scaling: StandardScaler was used to normalize input features.
• Model Evaluation: Performance was assessed using accuracy, ROC-AUC, precision, recall, and F1-score. A confusion matrix and feature importance analysis were also generated.

4. Web Application Deployment

• Tool: Streamlit was used to create an interactive web app.
• Features:
  • Input interface for SMILES strings.
  • Real-time descriptor calculation and bioactivity prediction.
  • Visualization of molecular structures and prediction confidence scores.

Results and Validation

• The trained XGBoost model achieved high accuracy in classifying COX-2 inhibitors.
• Validation was performed using known natural COX-2 inhibitors (e.g., Senkyunolide I, Cryptotanshinone, Roburic acid), and the model correctly predicted their activity.
• Key insights from feature importance analysis highlighted molecular descriptors critical for COX-2 inhibition.

Key Features

• Data-Driven Approach: Utilized experimentally validated bioactivity data from ChEMBL.
• Interpretability: Feature importance analysis provides insights into molecular properties influencing bioactivity.
• User-Friendly Interface: The Streamlit app enables researchers to make predictions without coding expertise.
• Scalability: The methodology can be adapted for other drug targets.

Future Work

• Expand Descriptors: Incorporate 3D molecular descriptors and docking scores for enhanced accuracy.
• Advanced Models: Explore deep learning techniques like graph neural networks for improved predictions.
• Experimental Validation: Collaborate with labs to test top predicted compounds in vitro.
• Multi-Target Predictions: Extend the model to predict activity against multiple biological targets.

How to Use

1. Clone the Repository:

   git clone https://github.com/AkhilT044/ML_Bioactivity_Pediction_CYCLOGENASE-2.git
   cd ML_Bioactivity_Pediction_CYCLOGENASE-2

2. Set Up the Environment:

   pip install -r requirements.txt

3. Run the Streamlit App:

   streamlit run app.py

4. Input SMILES Strings: Use the web interface to input SMILES strings of compounds and view predictions.