Predicting Longitudinal Traits Derived from High-Throughput Phenomics in Contrasting Environments Using Genomic Legendre Polynomials and B-Splines. Issue 10 (1st October 2019)

• Record Type: Journal Article
• Authors: Momen, Mehdi; Campbell, Malachy T; Walia, Harkamal; Morota, Gota
• Is Part Of: G3. Volume 9:Issue 10(2019)
• Page Start: 3369
• Page End: 3380
• Publication Date: 2019-10-01
• Subjects: genomic prediction -- phenomics -- longitudinal modeling -- random regression -- time series -- GenPred -- Shared Data Resources; Genetics -- Research -- Periodicals; Genomics -- Periodicals; Genetics; Genomics; Genes; Genetics -- Research; Genomics; Electronic journals; Periodical; Periodicals; Fulltext; Internet Resources; Periodicals; 572.8
• DOI: 10.1534/g3.119.400346
• Languages: English
• ISSNs: 2160-1836

Abstract

Abstract: Recent advancements in phenomics coupled with increased output from sequencing technologies can create the platform needed to rapidly increase abiotic stress tolerance of crops, which increasingly face productivity challenges due to climate change. In particular, high-throughput phenotyping (HTP) enables researchers to generate large-scale data with temporal resolution. Recently, a random regression model (RRM) was used to model a longitudinal rice projected shoot area (PSA) dataset in an optimal growth environment. However, the utility of RRM is still unknown for phenotypic trajectories obtained from stress environments. Here, we sought to apply RRM to forecast the rice PSA in control and water-limited conditions under various longitudinal cross-validation scenarios. To this end, genomic Legendre polynomials and B-spline basis functions were used to capture PSA trajectories. Prediction accuracy declined slightly for the water-limited plants compared to control plants. Overall, RRM delivered reasonable prediction performance and yielded better prediction than the baseline multi-trait model. The difference between the results obtained using Legendre polynomials and that using B-splines was small; however, the former yielded a higher prediction accuracy. Prediction accuracy for forecasting the last five time points was highest when the entire trajectory from earlier growth stages was used to train the basis functions. Our results suggested that it was possible to decrease phenotyping frequency by only phenotyping every other day in order to reduce costs while minimizing the loss of prediction accuracy. This is the first study showing that RRM could be used to model changes in growth over time under abiotic stress conditions.