|
Abstract:
|
Spatial transcriptomic studies perform gene expression profiling on tissues with spatial localization information. Here, we present a statistical method, BASS, for effective spatial transcriptomic analysis that examines the hierarchical organization of tissues at two distinct scales. Specifically, at the single-cell scale, BASS performs cell type clustering and clusters cells into cell types. At the tissue regional scale, BASS segments the tissue section into distinct spatial domains in a de novo fashion. Importantly, BASS performs both analyses in a coherent fashion through a Bayesian hierarchical modeling framework, allowing for seamless integration of gene expression information with spatial information to improve the analyses at both scales. Moreover, BASS can perform multi-sample analysis via joint modeling of multiple tissue samples, facilitating the cross-sample integration of spatial transcriptomics. We illustrate the benefits of BASS through comprehensive simulations and applications to three spatial transcriptomic datasets. The substantial power gain brought by BASS allowed us to reveal accurate transcriptomic and cellular landscape in both cortex and hypothalamus.
|
