A multi-trait Bayesian method for mapping QTL and genomic prediction

Kemper, Kathryn E., Bowman, Philip J., Hayes, Benjamin J., Visscher, Peter M. and Goddard, Michael E. (2018) A multi-trait Bayesian method for mapping QTL and genomic prediction. Genetics Selection Evolution, 50 . doi:10.1186/s12711-018-0377-y

Author Kemper, Kathryn E.
Bowman, Philip J.
Hayes, Benjamin J.
Visscher, Peter M.
Goddard, Michael E.
Title A multi-trait Bayesian method for mapping QTL and genomic prediction
Journal name Genetics Selection Evolution   Check publisher's open access policy
ISSN 1297-9686
Publication date 2018-03-24
Sub-type Article (original research)
DOI 10.1186/s12711-018-0377-y
Open Access Status DOI
Volume 50
Total pages 13
Place of publication London, United Kingdom
Publisher BioMed Central
Language eng
Formatted abstract
Background: Genomic prediction and quantitative trait loci (QTL) mapping typically analyze one trait at a time but this may ignore the possibility that one polymorphism affects multiple traits. The aim of this study was to develop a multivariate Bayesian approach that could be used for simultaneously elucidating genetic architecture, QTL mapping, and genomic prediction. Our approach uses information from multiple traits to divide markers into ‘unassociated’ (no association with any trait) and ‘associated’ (associated with one or more traits). The effect of associated markers is estimated independently for each trait to avoid the assumption that QTL effects follow a multi-variate normal distribution.

Results: Using simulated data, our multivariate method (BayesMV) detected a larger number of true QTL (with a posterior probability > 0.9) and increased the accuracy of genomic prediction compared to an equivalent univariate method (BayesR). With real data, accuracies of genomic prediction in validation sets for milk yield traits with high-density genotypes were approximately equal to those from equivalent single-trait methods. BayesMV tended to select a similar number of single nucleotide polymorphisms (SNPs) per trait for genomic prediction compared to BayesR (i.e. those with non-zero effects), but BayesR selected different sets of SNPs for each trait, whereas BayesMV selected a common set of SNPs across traits. Despite these two dramatically different estimates of genetic architecture (i.e. different SNPs affecting each trait vs. pleiotropic SNPs), both models indicated that 3000 to 4000 SNPs are associated with a trait. The BayesMV approach may be advantageous when the aim is to develop a low-density SNP chip that works well for a number of traits. SNPs for milk yield traits identified by BayesMV and BayesR were also found to be associated with detailed milk composition.

Conclusions: The BayesMV method simultaneously estimates the proportion of SNPs that are associated with a combination of traits. When applied to milk production traits, most of the identified SNPs were associated with all three traits (milk, fat and protein yield). BayesMV aims at exploiting pleiotropic QTL and selects a small number of SNPs that could be used to predict multiple traits.
Q-Index Code C1
Q-Index Status Provisional Code
Grant ID DP1093502
Institutional Status UQ

Version Filter Type
Citation counts: TR Web of Science Citation Count  Cited 0 times in Thomson Reuters Web of Science Article
Scopus Citation Count Cited 0 times in Scopus Article
Google Scholar Search Google Scholar
Created: Wed, 28 Mar 2018, 10:04:14 EST