BANKSY: A Spatial Omics Algorithm that Unifies Cell Type Clustering and Tissue Domain Segmentation

Abstract

Abstract Each cell type in a solid tissue has a characteristic transcriptome and spatial arrangement, both of which are observable using modern spatial omics assays. However, the common practice is still to ignore spatial information when clustering cells to identify cell types. In fact, spatial location is typically considered only when solving the related, but distinct, problem of demarcating tissue domains (which could include multiple cell types). We present BANKSY, an algorithm that unifies cell type clustering and domain segmentation by constructing a product space of cell and neighbourhood transcriptomes, representing cell state and microenvironment, respectively. BANKSY’s spatial kernel-based feature augmentation strategy improves per-formance and scalability on both tasks when tested on FISH-based and sequencing-based spatial omics data. Uniquely, BANKSY identified hitherto undetected niche-dependent cell states in two mouse brain regions. Lastly, we show that quality control of spatial omics data can be formulated as a domain identification problem and solved using BANKSY. BANKSY represents a biologically motivated, scalable, and versatile framework for analyzing spatial omics data.