James Schnable:
Learning to lead the target: Developing crops for changing environments


Bionote

James Schnable is an Associate Professor in the Department of Agronomy & Horticulture at the University of Nebraska Lincoln where his work focuses on maize and sorghum genomics, genetics, and genotype by environment interaction. Students in the Schnable Lab collaborate closely with faculty in statistics, computer science, or engineering. Schnable has helped found three startups commercializing bioinformatic, quantitative genetic, or digital agricultural technologies. His research is currently supported by the US Department of Agriculture, the National Science Foundation, the Department of Energy, and the Nebraska Corn Growers.

Presentation Abstract

Modern plant breeding has depended on data obtained from yield trials in the target environment to both identify desirable new crop varieties for release to farmers and to train genomic prediction models which can improve the efficiency and rate of genetic gain of future breeding cycles. Field trials conducted in one environment provide only extremely limited insight into which crop varieties will exhibit good performance in another environment. Current benchmark datasets for cross environment prediction are typically either generated by industry and lack rich meta-data associated with individual lines or include only modest numbers of lines evaluated in each environment. Genomic predictions can outperform real world trait data in accuracy of cross environment prediction and phenotypic prediction using non-linear models can outperform genomic prediction. In a preliminary analysis using the random forest algorithm to predict yields in the Lincoln, NE 2020 field site with all yield and ear related traits removed, permutation analysis suggested the three most important traits for prediction accuracy were: 1) Days to anthesis, 2) number of branches per tassel and 3) plant height. These three traits are all potentially scorable from UAV data, suggesting it may be possible to predict variation in end-season yields using high throughput phenotyping data collected during the growing season. Conducting linked yield trials across distinct environments combined with mining data from the literate enabled the deployment of cross environment prediction methodologies that quantify and leverage the different roles comparatively stable plant phenotypes play in determining plastic phenotypes (e.g. grain yield) across environments.