Pharm.D candidate · South Korea
Applying machine learning to drug discovery and biomarker data, and checking hard whether the results hold up.
- Cheminformatics: predicting drug–transporter interactions (P-gp/ABCB1) from molecular structure
- Bioinformatics: mining metagenomes for antimicrobial peptides, and classifying disease from miRNA profiles
- Validation-first ML: nested CV, paired fold-level comparisons, false-positive base rates, and confounder checks before claiming a result
| Project | What it is | Key result |
|---|---|---|
| P-gp inhibitor prediction | Literature-mined dataset of 522 compounds + ML/GNN benchmark | Random Forest (ROC-AUC 0.80) outperformed GCN, GAT, and GINE on identical folds |
| Polar AMP discovery | Antimicrobial peptide screening of the Arctic permafrost virome, with a 6-part reality check | Showed the "discoveries" were classifier noise · 2026 Polar Big Data & AI Competition |
| Exosomal miRNA × CRC | Colorectal cancer classification from serum exosomal miRNA (GSE39833) | Found that sample run order is perfectly confounded with the label · 🏆 1st prize, CHA University research poster (2025) |
| Caffeine–drug checker | Streamlit app that checks caffeine intake against current medications, with a PDF report | Early project (2024), my first code |
| Supplement chatbot | Symptom-based supplement suggestions; a first step toward a counseling app | Discontinued prototype (2025), stopped over health-information legal risk |
A result is only interesting after I've tried to break it.