Research
Virtual Screening Using a Novel Contrastive Learning Framework
Advisor: Prof. Jarek Meller
- Developing a contrastive learning framework for protein-family-aware virtual screening.
- Studying how such representations transfer across major protein target families.
Manuscript in preparation
Prediction of Opioid–Induced Respiratory Depression in Pediatric Patients
Advisor: Prof. Jarek Meller
- Developing machine-learning models to predict opioid-induced respiratory depression from genotype and clinical features in a multi-center pediatric cohort.
- Designing robust, consensus-based feature-selection methods for identifying variants associated with risk.
Manuscript in preparation
Benchmarking the Generalization of Chemical Language Models
Advisor: Prof. Jarek Meller
- Evaluated how well chemical language models (MolFormer, ChemBERTa) separate active from inactive compounds in embedding space, at both the target and the family level.
- Conclusion: training ML classifiers on top of the DUD-E benchmark is not appropriate; DUD-E is better used to benchmark the generalization of chemical language models directly.