Tensor decomposition for multiple-tissue gene expression experiments

Nat Genet. 2016 Sep;48(9):1094-100. doi: 10.1038/ng.3624. Epub 2016 Aug 1.

Abstract

Genome-wide association studies of gene expression traits and other cellular phenotypes have successfully identified links between genetic variation and biological processes. The majority of discoveries have uncovered cis-expression quantitative trait locus (eQTL) effects via mass univariate testing of SNPs against gene expression in single tissues. Here we present a Bayesian method for multiple-tissue experiments focusing on uncovering gene networks linked to genetic variation. Our method decomposes the 3D array (or tensor) of gene expression measurements into a set of latent components. We identify sparse gene networks that can then be tested for association against genetic variation across the genome. We apply our method to a data set of 845 individuals from the TwinsUK cohort with gene expression measured via RNA-seq analysis in adipose, lymphoblastoid cell lines (LCLs) and skin. We uncover several gene networks with a genetic basis and clear biological and statistical significance. Extensions of this approach will allow integration of different omics, environmental and phenotypic data sets.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Adipose Tissue / metabolism*
  • Bayes Theorem*
  • Cohort Studies
  • Gene Expression Profiling
  • Gene Expression Regulation
  • Gene Regulatory Networks
  • Genetic Markers / genetics*
  • Genome-Wide Association Study
  • Humans
  • Lymphocytes / metabolism*
  • Phenotype
  • Polymorphism, Single Nucleotide / genetics*
  • Quantitative Trait Loci
  • Skin / metabolism*
  • United Kingdom

Substances

  • Genetic Markers