Genotype ordination from factor analytic covariance models under missing genotype-by-environment data

BRUNO CECILIA; BALZARINI MÓNICA

Riga

Conferencia; 31st International Biometric Conference (IBC 2022); 2022

International Biometric Society

Additive main fixed effects multiplicative interaction (AMMI) models are using to exploit the genotype by environment interaction (GEI) in multi-environmental (E) trials to compare the performance of genotypes (G). Frequently, the trials comprise a group of G that are evaluated in all E (complete dataset) and the GEI is explored through principal component analysis and its representation by biplot (AMMI-biplots). Alternatively, GEI can be include as random effect using a linear mixed models (LMM) with factor analytic (FA) co-variance structure. Although this model can be fitted in context with an incomplete GEI dataset, the relative performance of G could be dependents on the total number of G and E evaluated, of the dimension of original combination GEI and/or the number of missing data. The aim of this work was to evaluate the ordination consensus among G obtained from AMMI-biplot with respect to the ordination form FA-biplot obtained under a complete dataset and with increased levels of the incompleteness data. We evaluated four complete datasets of Argentinean multi-environment yield trials in wheat with different dimensions of GEI (12x15, 15x18, 14x15, and 16x15). The E was defined as combination of sites and growing season among 2016 and 2018. Each trials was conducted under DBCA with three repetitions. The grade of incompleteness, from 5% to 50%, was reached by delete in each growing season and each site the G with the lowest performance. The consensus of ordination of remainder G with the G ordination in complete dataset was evaluated with Generalized Procrustes Analysis. We found a high consensus between G ordination obtained by AMMI-biplot with ordination by FA-biplot after a LMM and the standardized score of G and loading of E. The FA-biplot with incomplete dataset shows a statistical significant consensus among the G ordination that remain in the trial after remove other G. This performance were independently of the dimension and the unbalanced evaluated. The propose of visualized with a kind of biplot from LMM with random GEI and covariance FA model is robust in context with missing data in multi-environment yield trials.