Structural equation models for genome-wide association study in Coffea arabica

Abstract

Coffee breeding techniques were based on classical techniques for a long time, however, with the advent of genomic techniques and precision phenotyping, breeding programs have been showing best and faster results, even with the programs becoming more complex, in terms of quantities and types of characteristics studied. Thus, the existence of interrelationships between characters can generate important impacts in a breeding program, such as the discovery of genomic regions that contribute to certain characteristics, these can act directly, or indirectly. Knowing this, understanding the direct and indirect effects that one character has on another is of great importance for the selection phase. Traditionally, multivariate techniques are applied, but phenotypic interrelationships are neglected. Thus, the use of the Bayesian Network (BN) in conjunction with the Structured Equation Model (SEM) under the focus of the genomic wide association study (GWAS), allows quantifying genetic parameters, partitioning such values into direct and indirect effects for the traits. present in the formed network. In order to explore these interrelationships, they were able to phenotypes related to morphological (fruit size and number of reproductive nodes), physiological (vegetative vigor) and productive (production) characteristics in 195 Coffea arabica genotypes from a partnership between Empresa de Pesquisa Agropecuária de Minas Gerais (EPAMIG), Empresa Brasileira de Pesquisa Agropecuária (EMBRAPA) and Federal University of Viçosa (UFV). The phenotypic network inferred by means of the Hill Climbing algorithm was used to estimate the appropriate coefficients. By performing an integration between multivariate models - GWAS and SEM- GWAS it was possible to identify a positive interrelationship between vegetative vigor in yield and vegetative vigor for the number of reproductive nodes and negative for the number of reproductive nodes and fruit size for yield. It was also possible to detect significant genomic regions, and thus identify three genes that act directly on yield. Keywords: Coffea arabica. Bayesian Network. Structural Equation Models. GWAS.