Joint cell type identification in spatial transcriptomics and single-cell RNA sequencing data

Abstract

We present ST-Assign, a novel computational tool for joint cell-type annotation in single-cell RNA sequencing and cell-type mixture decomposition in spatial transcriptomics data. The model integrates the two data sources to enhance cell-type identification. It accounts for shared variables such as gene expression profiles and leverages prior knowledge about marker genes. We formulate the model as a generative graph-based structure and employ Markov chain Monte Carlo for the inference. We demonstrate the model's utility on mouse brain data, achieving simultaneous cell-type annotation and decomposition of cell-type mixtures. ST-Assign provides valuable insights into cell populations within organisms, improving our understanding of cellular heterogeneity.

Navigating the landscapes of spatial transcriptomics: How computational methods guide the way

“…Other methods can perform spatial deconvolution without a reference, relying only on spatial omics data. These methods can be either data-driven, such as STdeconvolve (Miller et al, 2022), a generative statistical model that uses latent Dirichlet allocation to infer cell types, or instructed by marker genes, such as Cellscope (Geras et al, 2023), ST-Assign (Agnieszka & Ewa, 2023), and SMART (Yang, Sin, & Ng, 2023). Prior knowledge of marker genes is usually obtained from public databases or literature.…”

“…Prior knowledge of marker genes is usually obtained from public databases or literature. For example, based on marker genes, ST-Assign uses a Metropoliswithin-Gibbs sampler to simultaneously perform cell type annotation and decomposition (Agnieszka & Ewa, 2023), whereas the SMART method uses a semi-supervised topic model to simultaneously infer the cell type composition and cell-type-specific gene expression profiles (Yang, Sin, & Ng, 2023).…”