arXiv (Cornell University) · 2019 · 63 citations · 0 references
Life SciencesBioinformaticsBiologyComputational BiologyGene ExpressionTranscriptomicsSingle-cell TranscriptomicsSpatial Gene ExpressionSpatial TranscriptomicsTranscriptomics IntegrationComputational Biology ModelsDomain AdaptationGene Expression ImputationSingle-cell Rna SequencingSparse Data ImputationSpatial Transcriptomic IntegrationSpatial Transcriptomics IntegrationTranscriptomic Data Integration
Spatial studies of transcriptome provide biologists with gene expression maps\nof heterogeneous and complex tissues. However, most experimental protocols for\nspatial transcriptomics suffer from the need to select beforehand a small\nfraction of genes to be quantified over the entire transcriptome. Standard\nsingle-cell RNA sequencing (scRNA-seq) is more prevalent, easier to implement\nand can in principle capture any gene but cannot recover the spatial location\nof the cells. In this manuscript, we focus on the problem of imputation of\nmissing genes in spatial transcriptomic data based on (unpaired) standard\nscRNA-seq data from the same biological tissue. Building upon domain adaptation\nwork, we propose gimVI, a deep generative model for the integration of spatial\ntranscriptomic data and scRNA-seq data that can be used to impute missing\ngenes. After describing our generative model and an inference procedure for it,\nwe compare gimVI to alternative methods from computational biology or domain\nadaptation on real datasets and outperform Seurat Anchors, Liger and CORAL to\nimpute held-out genes.\n