Research
My research connects representation learning, single-cell measurement, and experimental systems to make complex cellular behavior interpretable and testable.
First-author methodMachine learning for cell state
A self-supervised framework for integrating, denoising, and reducing single-cell data while preserving local geometry and global topology.
Dataset-aware and hard-negative sampling let a minimalist contrastive model resolve complex trajectories across batches, technologies, and species.
First-author methodStatistical genomics
A statistical method for recovering sample identity and doublets from multiplexed single-cell experiments with noisy or unbalanced tag counts.
Mechanistic contamination models, negative-binomial GLMs, and expectation–maximization retain more high-confidence singlets.
Co-first-author studyCancer systems biology
Single-cell maps, organoids, and immune-cell co-culture connect predicted colorectal carcinoma interactions to experimental mechanism.
Tumor cells instructed macrophages toward immunosuppressive, pro-tumorigenic states in a reductionist co-culture system.
First-author studyDevelopmental hematopoiesis
Single-cell transcriptomics resolved the endothelial-to-hematopoietic transition and the molecular progression toward pre-HSC identity.
A continuous endothelial-to-hematopoietic trajectory revealed a developmental bottleneck before hemogenic endothelium and two temporally distinct hematopoietic outputs.
Co-first-author studyDevelopmental genomics
A single-cell atlas connected transcriptional states to the invariant embryonic lineage, revealing how lineage history and cell fate organize development.
More than 86,000 single-cell transcriptomes were mapped onto the known lineage, reconstructing developmental trajectories for nearly every embryonic cell type.
