MoleculeRankAI

Explainable AI Platform for Drug Discovery

MoleculeRankAI is an AI-powered molecular ranking platform designed to prioritize promising drug candidates from large chemical datasets.

The system combines molecular feature engineering, machine learning models, ensemble ranking, and explainable AI to help researchers identify molecules with higher potential for laboratory testing.

Project Pipeline

Public Drug Data
        |
        v
Data Cleaning
        |
        v
Molecular Feature Generation
        |
        |
 -----------------------------
 |             |             |
Fingerprints  Descriptors   Graphs
 |             |             |
 -----------------------------
              |
              v
       AI Prediction Models

    Random Forest
    XGBoost
    Graph Neural Network

              |
              v

       Ensemble Ranking

              |
              v

       SHAP Explainability

              |
              v

       Scientist Dashboard

Features

• Molecular structure analysis
• Molecular property prediction
• Candidate ranking system
• Ensemble AI scoring
• SHAP-based explainability
• Interactive Streamlit dashboard

Technologies

• Python
• RDKit
• Scikit-learn
• XGBoost
• PyTorch
• PyTorch Geometric
• SHAP
• Streamlit

Project Structure

MoleculeRankAI/

├── data/
├── notebooks/
├── models/
├── src/
├── dashboard/
├── requirements.txt
└── README.md

Dashboard

Launch the application:

streamlit run dashboard/app2.py

The dashboard allows users to:

• Browse ranked molecules
• Search candidates
• Inspect molecular structures
• View prediction results
• Explore AI explanations

Future Improvements

• Additional ADMET prediction models
• Larger pharmaceutical datasets
• Improved graph neural network architectures
• Experimental validation workflows