40 Factors Influencing Accuracy of Genomic Selection with Sequence Information

Abstract

Abstract Incorporating the sequence information only marginally increases the accuracy of genomic selection. The purpose of this study was to find out why by examining profiles of Quantitative Trait Nucleotides (QTN). Multiple populations were simulated with different effective population sizes and number of animals. 100 equidistant QTN with identical substitution effects were included in 50k SNP genotypes. Analyses were by single-step GBLUP, with solutions converted to SNP values and subsequently to p-values for each SNP. Manhattan plots for standardized SNP solutions were noisy and were elevated only for few QTNs. Manhattan plots for p-values were similar to those for SNP solutions, indicating little impact of population structure. The number of significant QTN was lower with lower effective population size and increased with larger data; at most about 20% of QTNs were detected. A QTN profile was created by averaging SNP solutions ±100 SNP around each QTN. The profile showed a normal-like response but with a distinct peak for the QTN. While the peak was higher with more data and higher effective population size, the normal-like response was smaller with higher effective population size. QTNs explained little variance because of shrinkage. The accuracy of genomic selection would be 100% if all QTNs are identified and their variances known, to prevent shrinking or inflation. This study allows to see limits of application of QTN from sequence data for genomic selection. If all causative SNP are included in the data, only a fraction of them can be identified even under a very simplistic architecture. As variance of QTN are assumed constant or are crude approximations (like in BayesR), the estimated QTN effects are inaccurate. Additional complications in QTN detection are close-spaced QTN and false QTNs due to imputation. Small effective population size allows the genomic selection by GBLUP but complicates the use of QTNs.