QSAR Predictive Modeling

Overview

Led the development of Quantitative Structure-Activity Relationship (QSAR) models to predict the biological activity of molecules related to Alzheimer's disease. Using Deep Neural Networks with Keras 3 and Mutual Information for feature selection, the model achieved a Mean Absolute Percentage Error (MAPE) of 7.3% and a Q² Ext of 0.87 on Alzheimer's-related compounds. Demonstrated the framework's versatility by successfully adapting and validating the same architecture for antibiotic discovery with comparable performance.

Key Achievements

• Developed QSAR models using Deep Neural Networks (Keras 3) and Mutual Information for feature selection.
• Achieved MAPE of 5.52% and R² of 0.81 in K-Fold cross-validation for antibiotic-related molecules.
• Alzheimer's model reached MAPE of 7.3% and Q² Ext of 0.87 during external validation.
• Successfully presented findings at the internal research symposium of the Institute of Chemistry.
• Prepared a manuscript detailing methodology and results, in final stages of preparation for submission to arXiv.