Improvements of resolutions and spatial context in sequencing technologies over the past several years have deepen the understanding of complex biological systems. However, these improvements also pose substantial computational and statistical challenges when analyzing the high-dimensional data, including accurate identification of key biological features, such as functional spatial domains and cell type identities, and in leveraging this information to dissect bulk genomics data.

In this talk, the presenter will present three research projects that address these challenges. The first focuses on uncovering spatial domain dynamics in spatiotemporal transcriptomics. This work combines variational autoencoders with Gaussian processes to learn spatially-aware embeddings, and integrates optimal transport modeling to characterize domain transitions over time. The second project introduces Cellcano, a supervised learning framework for accurate and robust cell type identification in single-cell ATAC-seq data. By using entropy to select high-confidence anchor cells, Cellcano enhances cross-dataset prediction performance. The final project, LRcell, is a regression-based method that incorporates single-cell marker gene information to infer the sub-cell-type sources of differential expression signals from bulk RNA-seq data.