Data-driven assessment of eQTL mapping methods

Abstract

BackgroundThe analysis of expression quantitative trait loci (eQTL) is a potentially powerful way to detect transcriptional regulatory relationships at the genomic scale. However, eQTL data sets often go underexploited because legacy QTL methods are used to map the relationship between the expression trait and genotype. Often these methods are inappropriate for complex traits such as gene expression, particularly in the case of epistasis.ResultsHere we compare legacy QTL mapping methods with several modern multi-locus methods and evaluate their ability to produce eQTL that agree with independent external data in a systematic way. We found that the modern multi-locus methods (Random Forests, sparse partial least squares, lasso, and elastic net) clearly outperformed the legacy QTL methods (Haley-Knott regression and composite interval mapping) in terms of biological relevance of the mapped eQTL. In particular, we found that our new approach, based on Random Forests, showed superior performance among the multi-locus methods.ConclusionsBenchmarks based on the recapitulation of experimental findings provide valuable insight when selecting the appropriate eQTL mapping method. Our battery of tests suggests that Random Forests map eQTL that are more likely to be validated by independent data, when compared to competing multi-locus and legacy eQTL mapping methods.

Reconstructability of Epistatic Functions

“…A variety of other approaches have been applied to modeling epistasis in gene expression [23][24][25][26][27][28]. Comparing RA to other approaches is beyond the scope of this paper, and will be the subject of future studies, but advantages of RA over some other methods can be stated briefly: (i) RA can be used for both nominal data and continuous function applications; this allows one to work within a single mathematical/computational framework, while some other methods are specific to only one these two applications; (ii) RA has three levels of refinement -coarse (variable-based models without loops), fine (variable-based models with loops), and ultra-fine (state-based models); this allows one to move smoothly within a single framework from broad search, e.g., GWAS, involving very many variables to ultrafine analysis that focuses on only a few, while some other methods (e.g., [21]) are specific to one of these situations; (iii) RA explicitly considers the space of possible models in its use of hypergraphs, and is thus especially designed for exploratory searches, while some other methods are primarily confirmatory, and require the user to specify the models to be considered; (iv) Even where RA overlaps with -and to the extent of the overlap is obviously not superior to -other methods, it has distinctive features, so it complements these other methods (see, e.g., the discussion of RA vs. Bayesian networks and other graphical models in [2]).…”

A random forest approach to capture genetic effects in the presence of population structure

“…Recent work revealed that such single-locus association models are frequently insufficient to explain the heritable component of complex traits [10][11][12] . Indeed, quantitative traits that involve both linear additive and epistatic (non-additive) effects appear to be the rule rather than an exception.…”

“…In particular, approaches based on random bagging techniques 27 have gained considerable attention. Implementations such as random forest (RF) 28 have been shown to accurately capture epistatic effects 10,29,30 . However, all of these approaches-including RF-assume that correlations between genotype and phenotype are genuine and, unlike the LMM, do not explicitly correct for population structure or other confounding effects.…”

Gene Regulatory Network Inference from Systems Genetics Data Using Tree-Based Methods

“…Methods that study the regulatory relationships at a systems-level include approaches based on Bayesian networks (Zhu et al 2007;Li et al 2005;Vignes et al 2011), structural equation models (Li et al 2006;Liu et al 2008), and the orientation of the edges of an undirected network using genetic markers as causal anchors (Aten et al 2008;Chaibub Neto et al 2008). Random Forests have also been successfully used for expression quantitative trait loci (eQTL) mapping (Michaelson et al, 2010).…”