Team project for CMU's 02-620: ML pipeline predicting inflamed vs non-inflamed Crohn's disease biopsies from single-cell RNA-seq (49 pre-treatment biopsies, GEO: GSE282122). I led the XGBoost modelling: grouped cross-validation by patient to prevent leakage, training on both raw gene features and PCA-reduced features, with feature importance analysis. XGBoost outperformed the kNN baseline.
Skills: Python, XGBoost, scikit-learn, PCA, Pandas
Team project for CMU's 02-604: network-based drug repurposing pipeline for cardiovascular disease that integrates GWAS data, STRING protein-protein interactions, and DGIdb drug-target data to surface repurposable compounds. I formalized the computational problem, built the STRING PPI network, and designed the randomization test that validated hub connectivity. The pipeline recovered both established CVD drugs and non-obvious candidates like ribavirin.
Skills: Python, NetworkX, STRING, DGIdb, GWAS data
Team project for CMU's 03-713: Comparative epigenomics pipeline identifying conserved and species-specific OCRs between human and mouse pancreas, then linking them to GO biological processes, promoter/enhancer classes, and TF motifs. I built Task 3, partitioning OCRs into conserved vs species-specific sets using bedtools-based reciprocal overlap analysis on HAL liftover outputs. The resulting BED files fed all downstream functional analyses.
Skills: Shell scripting, bedtools, HAL, rGREAT, HOMER, BED files