Tumor evolution leaves measurable fingerprints across RNA, DNA, and chromatin. Our lab develops probabilistic models that convert those fingerprints into actionable biomarkers for immunotherapy and targeted treatment selection.

Topic modeling for tumor microenvironment states. Inspired by natural language processing, we decompose bulk and single-cell profiles into “topics” representing stromal remodeling, T-cell exhaustion, and metabolic reprogramming. The approach explains variation in checkpoint inhibitor response across CALGB/Alliance trials.

We deliver open-source software (scTopics), reproducible Nextflow pipelines, and interactive dashboards so collaborators can explore biomarker hypotheses in real time. Manuscripts are under review at Genome Biology and JCO Precision Oncology; preprints and workshop slides are linked below.