Postdoctoral researcher
I’m Jingyuan Hu, a biostatistician and computational biologist at Harvard T.H. Chan School of Public Health. My research uses mathematics, statistics, and geometry to understand high-dimensional single-cell data.
Projects
I am interested in learning the manifold structure of single-cell gene-expression data and using differential geometry to understand cellular dynamics. This work includes FlowMap, a geometry-preserving method for embedding and interpreting RNA velocity.
I am interested in how increasing dataset size relates to the amount of biological information that can be recovered. Using tools from random matrix theory, we develop scaling laws that characterize the behavior of representation learning in single-cell data.
I am interested in representation learning in biological foundation models. In particular, I study whether these models encode regulatory elements and their interactions within their learned embedding spaces.
Publications
bioRxiv preprint · doi:10.64898/2026.09.21.753192
Cell 184(9), 2471–2486.e20
Developmental Cell 56(8), 1100–1117.e9
Software
FlowMap embeds single-cell gene expression and RNA velocity data into a smooth low-dimensional geometry where cell-state flow can be visualized, compared, and analyzed.
My YouTube channel:
@JingyuanHu01I’m enthusiastic about learning new things—especially mathematics—and I enjoy sharing that process with others. Modern computational biology draws on many branches of mathematics, often intersecting with genomics in unexpected ways. I see mathematics as a practical set of tools, and I taught myself much of what I use through YouTube. That experience showed me how freely shared knowledge can help anyone build expertise and expand their understanding far beyond formal training.
Video series: