Assessment of parametric and non-parametric methods for prediction of quantitative traits with non-additive genetic architecture
Abdolreza Salehi, Maryam Bazrafshan, Rostam Abdollahi-Arpanahi
Abstract
Open-access reader
Abdolreza Salehi, Maryam Bazrafshan, Rostam Abdollahi-Arpanahi
Abstract
Open-access reader
Abstract Whole genome evaluation of quantitative traits using suitable statistical methods enables researchers to predict genomic breeding values (GEBVs) more accurately. Recent studies suggested that the ability of methods in terms of predictive performance may depend on the genetic architecture of traits. Therefore, when choosing a statistical method, it is essential to consider the genetic architecture of the target traits. Herein, the performance of parametric methods i.e. GBLUP and BayesB and non-parametric methods i.e. Bagging GBLUP and Random Forest (RF) were compared for traits with different genetic architecture. Three scenarios of genetic architecture, including purely Additive (Add), purely Epistasis (Epis) and Additive-Dominance-Epistasis (ADE) were considered. To this end, an animal genome composed of five chromosomes, each chromosome harboring 1000 SNPs and four QTL was simulated. Predictive accuracies in the first generation of testing set under Additive genetic architectures for GBLUP, BayesB, Baging GBLUP and RF were 0.639, 0.731, 0.633 and 0.548, respectively, and were 0.278, 0.330, 0.275 and 0.444 under purely Epistatic genetic architectures. Corresponding values for the Additive-Dominance-Epistatic structure also were 0.375, 0.448, 0.369 and 0.458, respectively. The results showed that genetic architecture has a great impact on prediction accuracy of genomic evaluation methods. When genetic architecture was purely Additive, parametric methods and Bagging GBLUP were better than RF, whereas under Epistatic and Additive-Dominance-Epistatic genetic architectures, RF delivered better predictive performance than the other statistical methods.
OpenAlex reports 6 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Abstract Whole genome evaluation of quantitative traits using suitable statistical methods enables researchers to predict genomic breeding values (GEBVs) more accurately. Recent studies suggested that the ability of methods in terms of predictive performance may depend on the genetic architecture of traits. Therefore, when choosing a statistical method, it is essential to consider the genetic architecture of the target traits. Herein, the performance of parametric methods i.e. GBLUP and BayesB and non-parametric methods i.e. Bagging GBLUP and Random Forest (RF) were compared for traits with different genetic architecture. Three scenarios of genetic architecture, including purely Additive (Add), purely Epistasis (Epis) and Additive-Dominance-Epistasis (ADE) were considered. To this end, an animal genome composed of five chromosomes, each chromosome harboring 1000 SNPs and four QTL was simulated. Predictive accuracies in the first generation of testing set under Additive genetic architectures for GBLUP, BayesB, Baging GBLUP and RF were 0.639, 0.731, 0.633 and 0.548, respectively, and were 0.278, 0.330, 0.275 and 0.444 under purely Epistatic genetic architectures. Corresponding values for the Additive-Dominance-Epistatic structure also were 0.375, 0.448, 0.369 and 0.458, respectively. The results showed that genetic architecture has a great impact on prediction accuracy of genomic evaluation methods. When genetic architecture was purely Additive, parametric methods and Bagging GBLUP were better than RF, whereas under Epistatic and Additive-Dominance-Epistatic genetic architectures, RF delivered better predictive performance than the other statistical methods.
Key concepts: Epistasis, Genetic architecture, Parametric statistics, Biology, Quantitative trait locus, Dominance (genetics), Statistics, Computer science