Robust Bayesian Integrative Modeling of Single Cell and Spatially Resolved Transcriptomics Data

A recent technology breakthrough in spatially resolved transcriptomics (SRT) has enabled comprehensive molecular characterization at the cellular level while preserving spatial information. However, many SRT technologies (e.g., spatial transcriptomics) cannot achieve single-cell resolution. Instead, they measure the average gene expression of a mixture of cells within a spot arranged on a lattice. Here, we introduce iMOSCATO, a fully Bayesian model that integrates single-cell RNA sequencing (scRNA-seq) data and SRT data to simultaneously decompose cell-type compositions of spots and identify the underlying spatial domains. We incorporate the lattice structure by employing a Markov random field prior, improving the accuracy of both cell-type deconvolution and spatial domain detection. Moreover, we use a zero-inflated Dirichlet distribution to capture cell-type sparsity. Finally, iMOSCATO shows competitive performance in accuracy compared to existing methods in both simulation studies and two real data applications.