Non-negative spatial matrix factorization for multi-sample spatial transcriptomics data

Author(s): Yi Wang,Kasper Daniel Hansen

Affiliation(s): Department of Biostatistics, Johns Hopkins University

Social media: https://twitter.com/Yi_Wang___

Spatial resolved transcriptomics opens the door to analyzing gene expression data in the context of spatial position. Recently, Townes and Engelhardt developed NSF, a non-negative spatial matrix factorization method. NSF can be used to identify spatial-dependent latent factors that are associated with functional anatomical regions. However, the model formulation is currently limited to a single sample. However, an increasing number of spatial datasets contain multiple samples from different tissue sections. Properly modeled, such datasets could potentially allow us to identify matched regions across samples and increase the confidence in downstream analysis. However, there are few existing analysis methods for spatial data which allow joint analysis of multiple samples. Here, we extend NSF to allow for multi-sample analysis. Our method, which we term mNSF, identifies factors which are common across multiple samples, and does not depend on cross-sample spot alignment, which is currently a challenging problem. In simulations, our method is capable of identifying spatial structures which are rotated between samples, as well as structures which only appear in some, but not all, samples. Using real data, we show our method is capable of identifying matched functional regions in multi-sample spatial transcriptomics data.