Statistical sampling of missing environmental variables improves biophysical genomic prediction in wheat

Area of Science:

• Agricultural Science
• Genetics
• Plant Breeding

Background:

• Whole-genome prediction (WGP) has advanced crop breeding by utilizing extensive reference populations.
• Crop growth models (CGMs) require detailed environmental data, limiting their application with historical WGP data lacking such inputs.
• The CGM-WGP model integrates CGMs and WGP but faces accuracy challenges with incomplete environmental records.

Purpose of the Study:

• To investigate the efficacy of approximating missing environmental variables within the CGM-WGP framework.
• To enhance the prediction accuracy of crop traits, particularly grain yield, using historical breeding data.
• To assess the impact of imputing specific environmental variables on model performance.

Main Methods:

• Developed and applied the CGM-WGP algorithm to wheat data.
• Investigated the imputation of initial soil water content (InitlSoilWCont) and initial nitrate profile.
• Compared prediction accuracies using sampled versus actual environmental variables.

Main Results:

• Sampling initial soil water content (InitlSoilWCont) alone significantly improved prediction accuracy for grain number (0.07), yield (0.06), and protein content (0.03).
• Imputing InitlSoilWCont increased the average narrow-sense heritability of genotype-specific parameters by 0.05.
• Root mean square errors for grain number and yield were reduced by 7% for CGM and 31% for CGM-WGP using sampled InitlSoilWCont.

Conclusions:

• Approximating missing environmental variables, specifically initial soil water content, is a viable strategy to enhance CGM-WGP accuracy.
• This approach expands the utility of WGP by enabling the inclusion of historical datasets with incomplete environmental records.
• The findings demonstrate a significant advantage in utilizing imputation methods for environmental data in genomic prediction models.