Modeling spatial trends and enhancing genetic selection: An approach to soybean seed composition breeding

Abstract

AbstractSpatial variation is a recurrent issue in field trials and can cause obstacles in terms of genetic selection. Analyses that account for spatial variation within location can lead breeders to predict genetic values accurately across locations in multi‐environment trials (METs). The present study aims to fit spatial models for analyzing soybean [Glycine max (L.) Merr.] seed composition traits using a two‐stage analysis pipeline and to assess its efficiency relative to a single‐stage analysis setting. Seed protein content (SPC), seed oil content (SOC), and seed storage protein content (SSP) data were collected from 283 soybean genotypes tested in four environments (C1, C2, V1, and V2). In Stage 1 of the two‐stage analysis, a randomized complete block (RCB) design model as well as four two‐dimensional first‐order (AR1 ⊗ AR1) spatial models were fit in each dataset to determine the most suitable model for genetic prediction. Predicted genetic values were used as input data for Stage 2. The most used spatial model [5] in Stage 1 of this study had accommodated local and global residuals. The autocorrelation estimates depicted spatial trends, especially in terms of rows, while column autocorrelation coefficients were low for C1 and C2 because of the limited number of blocks and their short length. Broad‐sense heritability, mean accuracy, and selection gains were greater for all traits in the two‐stage analysis than in the single‐stage analysis. The two‐stage analysis leveraged the spatial model fitting in the Stage 1 and proved to be advantageous for soybean seed composition breeding.