Selective Shrinkage of Genomic Effects Using Synthetic Dependencies in Neighboring Chromosome Regions

Abstract Text: As the number of model parameters increases with still growing number of SNPs, multicollinearity between covariates can affect the results of whole genome prediction. Selecting appropriate SNPs may counteract this phenomenon. Additionally, dependencies between single SNPs or chromosome regions can directly be incorporated in prediction methods. In this study, relationships between regions were modelled synthetically via a base-function approach; the genetic effect at some locus is a linear combination of base-function effects referring to the underlying and preceding regions. This B-spline approach was combined with a stochastic variable selection method to identify regions with non-zero impact. Application to milk performance data of 1,295 Holstein cows showed little reduced estimates of genetic variance components with decreased standard error and drastically reduced computing time compared to the analysis including all SNPs. Due to the synthetic structure of dependencies, this approach is applicable to different species.

Keywords: SNP selection, model complexity, B-spline