HEFT: eQTL analysis of many thousands of expressed genes while simultaneously controlling for hidden factors.

Publication TypeJournal Article
Year of Publication2014
AuthorsGao C, Tignor NL, Salit J, Strulovici-Barel Y, Hackett NR, Crystal RG, Mezey JG
Date Published2014 Feb 01
KeywordsGene Expression, Gene Expression Profiling, Humans, Lung, Polymorphism, Single Nucleotide, Quantitative Trait Loci, Regression Analysis, Software

MOTIVATION: Identification of expression Quantitative Trait Loci (eQTL), the genetic loci that contribute to heritable variation in gene expression, can be obstructed by factors that produce variation in expression profiles if these factors are unmeasured or hidden from direct analysis.

METHODS: We have developed a method for Hidden Expression Factor analysis (HEFT) that identifies individual and pleiotropic effects of eQTL in the presence of hidden factors. The HEFT model is a combined multivariate regression and factor analysis, where the complete likelihood of the model is used to derive a ridge estimator for simultaneous factor learning and detection of eQTL. HEFT requires no pre-estimation of hidden factor effects; it provides P-values and is extremely fast, requiring just a few hours to complete an eQTL analysis of thousands of expression variables when analyzing hundreds of thousands of single nucleotide polymorphisms on a standard 8 core 2.6 G desktop.

RESULTS: By analyzing simulated data, we demonstrate that HEFT can correct for an unknown number of hidden factors and significantly outperforms all related hidden factor methods for eQTL analysis when there are eQTL with univariate and multivariate (pleiotropic) effects. To demonstrate a real-world application, we applied HEFT to identify eQTL affecting gene expression in the human lung for a study that included presumptive hidden factors. HEFT identified all of the cis-eQTL found by other hidden factor methods and 91 additional cis-eQTL. HEFT also identified a number of eQTLs with direct relevance to lung disease that could not be found without a hidden factor analysis, including cis-eQTL for GTF2H1 and MTRR, genes that have been independently associated with lung cancer.

AVAILABILITY: Software is available at http://mezeylab.cb.bscb.cornell.edu/Software.aspx.

SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Alternate JournalBioinformatics
PubMed ID24307700
PubMed Central IDPMC3904522
Grant ListP50 HL084936 / HL / NHLBI NIH HHS / United States
R01 HL074326 / HL / NHLBI NIH HHS / United States