Premium
Genomic prediction enables early but low‐intensity selection in soybean segregating progenies
Author(s) -
Mendonça Leandro de Freitas,
Galli Giovanni,
Malone Gaspar,
FritscheNeto Roberto
Publication year - 2020
Publication title -
crop science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.76
H-Index - 147
eISSN - 1435-0653
pISSN - 0011-183X
DOI - 10.1002/csc2.20072
Subject(s) - biology , germplasm , selection (genetic algorithm) , genomic selection , population , plant breeding , breeding program , microbiology and biotechnology , genotype , agronomy , genetics , single nucleotide polymorphism , cultivar , gene , machine learning , computer science , demography , sociology
In soybean [ Glycine max (L.) Merr.], new commercial lines are commonly obtained from biparental crosses, and the selection is performed as homozygosity increases. However, it is difficult to select for quantitative traits in the early steps of breeding, due to the high heterozygosity level and a vast number of new progenies, which sometimes lead breeders to randomly select for these traits in this phase. Therefore, we aimed to assess the impact of genomic selection in early generations of a soybean breeding program. Working on germplasm derived from two different maturity regions in Brazil, genotyped in F 2 and phenotyped in F 2:4 for grain yield, plant height, maturity rating, and days to maturity, we compared the composition of different training populations, models with and without the genotype × environment (G × E) interaction effect, and two types of relationship measurements (genetic similarity and Euclidian distance). Results showed superior performance of the Euclidian distance kernel over the standard VanRaden kernel in major scenarios tested. In general, G × E models did not obtain superior performance compared with mean principal models, and the training population composed only of the nearest progenies had the highest prediction ability. The best models achieved prediction abilities between 0.40 and 0.56, thereby enabling application of a low‐intensity selection in F 2 . As a result, half of the progenies could be discarded without missing a great part of the good ones. Our results show that through genomic prediction, it is possible to select for quantitative traits in the early steps of breeding, which might increase the efficiency of the program in the advanced phases.