Academic journal article Genetics

Bayesian Modeling of Haplotype Effects in Multiparent Populations

Academic journal article Genetics

Bayesian Modeling of Haplotype Effects in Multiparent Populations

Article excerpt

(ProQuest: ... denotes formulae omitted.)

ANincreasinglywell-establishedparadigmforinvesti- gating the genetics of complex traits is the multiparent cross (Churchill et al. 2004; Cavanagh et al. 2008; Kover et al. 2009; Johannesson et al. 2009; Huang et al. 2012; King et al. 2012a; Svenson et al. 2012; Bandillo et al. 2013; Mackay et al. 2014), and an increasingly popular target for genetic and phenotypic analysis is its result, the multiparent popula- tion (e.g.,Talbotet al. 1999; Valdar et al. 2006a,b; Huang et al. 2009; Aylor et al. 2011; Huang et al. 2011; Collaborative Cross Consortium 2012; Baud et al. 2013; Marriage et al. 2014; Tsaih et al. 2014). In a multiparent cross, a select set of known founders is combined and bred to an advanced generation to create a population of individuals whose genomes are mosaics of the original founder haplotypes; this multiparent population is then well suited for detection of quantitative trait loci (QTL) mapping through linkage disequilibrium (LD) mapping-that is, QTL mapping based on inferred descent. The development of statistical methods for LD mapping in these populations has largely focused on QTL detection (e.g.,Mottet al. 2000; Valdar et al. 2009; Durrant and Mott 2010; Huang and George 2011; Yuan et al. 2011; Zhang et al. 2012). Methods to characterize the effects at detected QTL, however, namely those estimat- ing how inheritance of alternate founder haplotypes drives phenotypic outcome (a key step in the design of follow-up studies), are relatively underdeveloped. This is particularly so for populations in which the identity of the haplotypes at the QTL is probabilistically inferred, where estimation of haplotype effects must proceed despite the fact that hap- lotype composition is itself uncertain.

Uncertainty in haplotype composition arises because hap- lotypes are not themselves the direct target of genotyping or sequencing assays. In a multiparent population, where the number of founders is J . 2, the markers used to genotype individuals will not be fully informative for descent at all (or often any) loci, and so the underlying haplotype mosaic of each individual must be inferred. Inference of the haplotype mosaic is typically performed using a suitably constructed hidden Markov model (HMM) (e.g.,Mottet al. 2000; Liu et al. 2010; King et al. 2012b; Gatti et al. 2014) and produces for each individual at each locus a list of probabilities describ- ing likely descent. For a diploid organism, this list enumerates the posterior probability of having inherited each haplotype pair (diplotype), given available genotype information on the individual and its founders.

Use of inferred haplotype composition rather than ob- served genotypes when testing for genetic association con- fers important advantages, including: automatic modeling of all ungenotyped, incompletely determined, or entirely uncharacterized genetic variants; increased robustness to genotyping errors due to borrowing of information across markers; implicit modeling of LD decay, leading to a clearer picture of available mapping resolution; and automatic modeling of local epistasis. This last benefit arises from the fact that the heritable range of allelic combinations of all variants within a QTL region is concisely circumscribed by local haplotype descent; this provides a more comprehen- sive account of genetic variation and thereby allows QTL to be detected with increased power.

Strictly speaking, QTL mapping based on inferred hap- lotypes should account for haplotype uncertainty (Lander and Botstein 1989), an idealized framework inferring both association and composition simultaneously (Lin and Zeng 2006). For large-scale detection of QTL, however, such com- bined inference is usually computationally impractical; it is also largely unnecessary given the existence of fast and pow- erful regression-based alternatives. Specifically, rapid and highly flexible detection of QTL canbeachievedbytreat- ing the inferred diplotype probabilities, or transformations thereof, as fixed covariates in an ordinary regression (Haley and Knott 1992; Mott et al. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.