Speaker:
Didong Li, PhD
Abstract: Spatial transcriptomics (ST) has emerged as a powerful tool, enabling the study of gene expression within tissue context. Numerous computational methods have been developed to analyze ST data, including spatial variable gene detection, spatial clustering, and deconvolution. However, these methods have mostly been designed and validated on small datasets, often limited to a single slide. One key limitation has been the lack of large-scale, publicly available ST datasets that integrate histopathology and pathologist annotations. To address this, we recently curated STimage-1K4M, a large-scale dataset featuring 1,149 slides, over 4 million image spot-gene expression pairs, across 50 tissue types and 8 species, with 71 slides containing pathologists' annotations. This opens doors for large-scale analysis using both statistical methods and modern approaches like large vision-language models. In this presentation, I will discuss some recently work based on this dataset, and explore the opportunities and challenges presented by large-scale ST data, highlighting the potential for new insights and methodological advancements.
Bio: Dr. Li is an Assistant Professor of Biostatistics, and an adjunct professor of Statistics and Operations Research, and Applied Math at UNC Chapel Hill. He got his PhD in Mathematics from Duke, followed by a postdoc training at Princeton Computer Science and UCLA Biostatistics. His research focus is statistical and machine learning methods development for robust inference of complex and high-dimensional data, specifically covering manifold learning, nonparametric Bayes, information geometry, and spatial statistics.
Zoom link: https://duke.zoom.us/j/97045144670