Epistatic Models Improve between Year Prediction and Prediction of Testcross Performance in Corn

Abstract

In a previous paper, the utility of partial least squares as a tool for predicting performance using marker‐based models was demonstrated. Including interactions between markers in prediction models improved prediction efficacy. The objectives of this paper were to determine whether: (i) molecular marker models based on 1 yr's phenotypic data could be used to predict performance in a second year; (ii) models based on per se data would be useful in predicting testcross performance; and (iii) adding epistasis to a model would improve prediction in either case. Data for protein, oil, and starch were obtained from 500 S2 lines and their testcrosses from the crosses of Illinois High Oil (IHO) × Illinois Low Oil (ILO) and of Illinois High Protein (IHP) × Illinois Low Protein (ILP) corn (Zea mays L.) strains. Adding epistasis to a model significantly increased predictive power both between years and for testcross performance. The proportion of variability accounted for when predicting testcross performance from per se performance was lower than when predicting performance in different years. In all cases, observed vs. predicted correlations were high enough to suggest they would be useful in marker‐assisted, or marker‐based, breeding.