Gene-Based Crop Modeling for Predicting Rice Phenological Variation across Multiple Environments.

Accurate prediction of phenotypic variation from genomic information is an important and major challenge to advance crop breeding. This research explored the feasibility of integrating crop models with genomic data to improve the prediction accuracy. A rice diversity panel with169 accessions was used in the study, with each accession consisting of 700,000 SNPs. Two rice models (ORYZA v3 and RicePSM) were used to predict phenotype data for days to 50% flowering (DF) in 10 environments. Of the 169 accessions, 135 were used as a training dataset and the remaining accessions were used as a validation dataset.

A total of 19 SNPs were first identified as potential genomic sites controlling days to flowering through a mixed linear model analysis of GWAS. Rice models were used to develop the crop parameters for phenological development up to 50% flowering using the phenotyping data of the training dataset. Relationships between crop parameters and the 19 SNP markers were derived with multiple linear regression using data in the training dataset. The relationships were then applied to estimate crop parameters of each accession in the validation dataset, and the parameters were fed into crop models to predict the DF values of each accession across the 10 environments. Results from both crop models showed that squared correlations (R²) between predicted and observed DF values varied between 0.58 and 0.82 among accessions while the relative root mean square error (RMSE_n) varied from 1.0 to 19.8%. The R² and RMSE_n for DF variations averaged across models, accessions and environments were 0.68 and 8.4%, much better than the reported results of pure statistical prediction. This research demonstrates that integration of crop models with genomic data can significantly improve the predictability of phenotypic variation across multiple environments and outperform statistical-based prediction.