Bayesian method to predict individual SNP genotypes from gene expression data

• Published in: Nature genetics Vol. 44; no. 5; pp. 603 - 608
• Main Authors: Schadt, Eric E, Woo, Sangsoon, Hao, Ke
• Format: Journal Article
• Language: English
• Published: New York Nature Publishing Group US 01.05.2012; Nature Publishing Group
• Subjects: 631/1647/2017; 631/208/199; 631/208/729/743; 631/553/2714; Adipose Tissue - metabolism; Agriculture; Animal Genetics and Genomics; Bayes Theorem; Bayesian statistical decision theory; Biological and medical sciences; Biomedical and Life Sciences; Biomedicine; Cancer Research; Cohort Studies; Computer Simulation; Deoxyribonucleic acid; DNA; Fundamental and applied biological sciences. Psychology; Gene expression; Gene Expression Profiling; Gene Function; Gene mapping; Genealogy; Genetics; Genetics of eukaryotes. Biological and molecular evolution; Genotype; Genotype & phenotype; Genotypes; Human Genetics; Humans; Identification and classification; Liver - metabolism; Methods; Polymorphism, Single Nucleotide - genetics; Privacy; Proteins; Quantitative Trait Loci; Single nucleotide polymorphisms; Studies; technical-report; United States; Genotype; Method; Single nucleotide polymorphism; Gene expression
• DOI: 10.1038/ng.2248
• ISSN: 1061-4036; 1546-1718

Eric Schadt and colleagues report a Bayesian method to predict individual SNP genotypes based on RNA expression data. Using simulations and empirical data sets, they show that it is possible to infer a genotypic barcode specific to an individual, although the identification of an individual as a participant in a study is limited by factors such as the availability of large-scale expression quantitative trait loci (eQTLs) and expression data sets. RNA profiling can be used to capture the expression patterns of many genes that are associated with expression quantitative trait loci (eQTLs). Employing published putative cis eQTLs, we developed a Bayesian approach to predict SNP genotypes that is based only on RNA expression data. We show that predicted genotypes can accurately and uniquely identify individuals in large populations. When inferring genotypes from an expression data set using eQTLs of the same tissue type (but from an independent cohort), we were able to resolve 99% of the identities of individuals in the cohort at P adjusted ≤ 1 × 10 −5 . When eQTLs derived from one tissue were used to predict genotypes using expression data from a different tissue, the identities of 90% of the study subjects could be resolved at P adjusted ≤ 1 × 10 −5 . We discuss the implications of deriving genotypic information from RNA data deposited in the public domain.