1. Active learning based Conformal Predictor, 07/01/23, Present, Designed a machine learning-assisted system using CatBoost and active learning to enhance virtual screening for ultra-large chemical libraries., Improved screening efficiency by 40% and reduced false positives in compound selection., Retrieved high-confidence compounds with minimal docking scores, streamlining drug discovery workflows.
2. Machine Learning Assisted Virtual Screening of Ultra large scale chemical library, 08/01/22, 06/30/23, Developed a Deep Learning-based solution using DMPNN (Deep Message Passing Neural Network) to identify ligands for IRAP (Insulin-Regulated Aminopeptidase)., Successfully identified the IRAP inhibitor as a starting point for further drug discovery.
3. Credit Card Fraud Detection, Designed a machine learning pipeline using SMOTE and Random Forest to predict fraudulent transactions., Achieved a high model performance with an AUC score of 0.98, improving detection accuracy for banks and consumers.