Prediction of count phenotypes using high-resolution images and genomic data

Area of Science:

• Agricultural Science
• Plant Breeding
• Genomics
• Machine Learning

Background:

• Genomic selection (GS) utilizes statistical machine learning for plant breeding.
• Traditional GS models face challenges with noisy phenotypic data.
• Integrating diverse data sources is crucial for enhancing prediction accuracy.

Purpose of the Study:

• To explore generalized Poisson regression (GPR) for genome-enabled prediction of count phenotypes.
• To investigate the integration of genomic data, environmental factors, and hyperspectral images in prediction models.
• To determine the optimal combination of data sources for improved breeding value prediction.

Main Methods:

• Developed a generalized Poisson regression (GPR) model.
• Integrated genotypic, environmental, and high-resolution hyperspectral image data.
• Analyzed interaction terms between data sources.

Main Results:

• The GPR model successfully integrated multiple data sources for prediction.
• Optimal prediction performance was achieved when genomic, environmental, and image data were combined.
• High-resolution image data acquired close to harvest yielded the best prediction results.

Conclusions:

• Generalized Poisson regression is a viable method for integrating diverse data in genomic selection.
• Combining genomic, environmental, and high-resolution image data significantly enhances prediction accuracy for plant breeding.
• Timing of image data acquisition (near harvest) is critical for maximizing predictive power.