Stat Cafe - Chen Yang
FlashDeconv: Atlas-scale Spatial Deconvolution via Structure-Preserving Sketching
- Time: Wednesday, 4/1/2026 from 11:30AM to 1:00PM
- Location: BLOC 503
Description
Spatial transcriptomics technologies are scaling to millions of spots, but existing deconvolution methods face computational bottlenecks. I’ll introduce FlashDeconv, which uses leverage-score importance sampling and randomized sketching to achieve linear O(N) complexity—processing 1 million spots in ~3 minutes without GPU. I’ll also discuss how this efficiency enables discovery of resolution-dependent artifacts in high-resolution platforms like Visium HD.
Our Speaker
Chen Yang is a second-year PhD student in Statistics at Texas A&M University. His research focuses on leveraging agentic AI for omics data analysis, particularly in single-cell and spatial transcriptomics. He develops computational tools for reproducible analysis of biological systems, including FlashDeconv, a scalable spatial deconvolution method for atlas-scale data.