FIST-nD: A tool for n-dimensional spatial transcriptomics data imputation via graph-regularized tensor completion

Abstract

Functional interpretation of spatial transcriptomics data usually requires non-trivial preprocessing steps and other supporting data in the analysis due to the high sparsity and incompleteness of spatial RNA profiling, especially in 3D constructions. As a solution, we present a new software tool FIST-nD, Fast Imputation of Spatially-resolved transcriptomes by graph-regularized Tensor completion in n-Dimensions for imputing 3D as well as 2D spatial transcriptomics data. FIST-nD is implemented based on a novel graph-regularized tensor decomposition method, which imputes spatial gene expression data using 4-way high-order tensor structure and relations in spatial and gene functional graphs. The implementation, accelerated by GPU or multicore parallel computing, can efficiently impute high-resolution 3D spatial transcriptomics data within a few minutes. The experiments on three 3D Spatial Transcriptomics datasets and one 3D high-resolution Stereo-seq dataset confirm the high accuracy of the imputation by FIST-nD and demonstrate that the imputed spatial transcriptomes provide a more complete gene expression landscape for downstream analyses such as spatial gene expression clustering and visualizations.