Single-cell RNA- and ATAC-seq profile tissues at cellular resolution, but the data are sparse and noisy: clustering smooths away meaningful heterogeneity, while pseudobulk loses fine structure.

With Sitara Persad and the lab of Dana Pe'er, we developed SEACells, a graph-based method that aggregates single cells into metacells — biologically coherent, high-resolution states that overcome sparsity while preserving heterogeneity. Metacells improve gene–peak association, regulatory analysis, and cross-modality integration, and have become a standard intermediate representation between single cells and clusters for large single-cell atlases.

Applying single-cell representations to cancer, we helped show that transcriptomic plasticity is a hallmark of metastatic pancreatic cancer — tumor cells shift their expression programs as they seed and adapt to new organs, a single-cell view of how metastases diversify.