APPLICATIONS OF MULTITRAIT AND MULTI-KERNEL MODELS FOR GENOMIC SELECTION IN AFRICAN CASSAVA.

Abstract

Genomic selection (GS) could help accelerate African cassava breeding towards the development of high yielding, high dry matter (DM), disease resistant and provitamin A varieties. This work addresses some issues for implementing GS in cassava. First, we evaluated multivariate and univariate GS models via prediction accuracies. Second, the genetic basis for DM content was investigated using the Regional Heritability Mapping (RHM) procedure. Lastly, the genetic basis for co-inheritance of DM, root color and fresh yield (FYLD) were investigated using the Regional co-heritability Mapping (RHM) procedure. Key lessons were: (1) Multitrait (MT) models for single location data offered 40% higher average prediction accuracies for genomic breeding values (GEBVs) of six target traits across 3 locations compared to single-trait (uT) models. (2) Multivariate multi-environment (ME) models also offered 12% higher average prediction accuracies compared to a compound symmetric multi-environment model (uE) parameterized as a univariate multi-kernel model for multi-year multi-environment data. (3) The RHM analysis identified segments associated with DM in white cassava on chromosomes 1, 4, 5, 10, 17,18 and on yellow cassava chromosome 1. Candidates extracted from genes adjacent to the RHM significant segments include: glycosyltransferases, serine-threonine kinases (SnRKs), invertases and fructose bisphosphate aldolase. Prediction accuracies from these candidates and all genes in the RHM significant regions suggest that they may be tagging regions associated with DM. (4) Genome-wide segment correlations from the RcHM analysis in yellow cassava showed a limited prospect for high DM yellow cassava development but good prospects for high DM, high yielding white cassava development.