Bayesian Joint Modeling of Single-Cell Expression Data and Bulk Spatial Transcriptomic Data

Single-cell RNA-sequencing (scRNA-seq) enables gene expression profiling at single-cell resolution, but it loses the spatial information of cells for solid tissues during the tissue dissociation step before sequencing. In contrast, bulk spatial transcriptomics (ST) methods can measure the expression...

Full description

Saved in:
Bibliographic Details
Published inStatistics in biosciences Vol. 15; no. 3; pp. 719 - 733
Main Authors Yu, Jinge, Wu, Qiuyu, Luo, Xiangyu
Format Journal Article
LanguageEnglish
Published New York Springer US 01.12.2023
Springer Nature B.V
Subjects
Adenocarcinoma
Algorithms
Bayesian analysis
Biostatistics
Gene expression
Gene sequencing
Health Sciences
Heterogeneity
Markov chains
Mathematics and Statistics
Medicine
Spatial data
Spatial heterogeneity
Statistical methods
Statistics
Statistics for Life Sciences
Theoretical Ecology/Statistics
Transcriptomics
Integrative analysis
Heterogeneity
scRNA-seq
Deconvolution
Spatial transcriptomics
Online AccessGet full text

Cover

More Information
Summary:Single-cell RNA-sequencing (scRNA-seq) enables gene expression profiling at single-cell resolution, but it loses the spatial information of cells for solid tissues during the tissue dissociation step before sequencing. In contrast, bulk spatial transcriptomics (ST) methods can measure the expression of spatially organized spots in solid tissues, but as a spot comprises dozens of cells, ST expression levels are averaged signals and lack cellular resolution. Joint analysis of these two complementary data types provides the opportunity to recover the spatial patterns of cell types and obtain the cellular enrichment of spots. However, there is a lack of unified statistical methods to achieve this goal. This study develops a Bayesian statistical method named BEATS to jointly model scRNA-seq data and bulk ST data from a common sample in the presence of cellular and spatial heterogeneity. BEATS can simultaneously (a) discover cell types, where cells in a cell type share mean expression profiles; (b) identify spot regions, where a region is a set of spots with the same cellular compositions; and (c) estimate cell-type proportions for each spot region. The Bayesian posterior inference is performed through a hybrid Markov chain Monte Carlo sampling algorithm. Extensive simulation studies and application to datasets on pancreatic ductal adenocarcinoma tissues demonstrate the practical utility of BEATS.
ISSN:1867-1764
1867-1772
DOI:10.1007/s12561-021-09308-4