Regularized selection indices for breeding value prediction using hyper-spectral image data.

Area of Science:

• Agricultural Science
• Genetics
• Data Science

Background:

• High-throughput phenotyping (HTP) generates vast amounts of data for crop improvement.
• Integrating high-dimensional HTP data into genetic analyses and breeding schemes presents statistical and computational challenges.

Purpose of the Study:

• To develop a novel methodology, regularized selection indices (SIs), for effectively utilizing high-dimensional HTP data in breeding.
• To enhance the accuracy of predicting economically relevant traits like grain yield.

Main Methods:

• Developed regularized selection indices by integrating penalization and rank-reduction techniques into the traditional SI framework.
• Applied the methodology to extensive wheat breeding data from the International Maize and Wheat Improvement Center (CIMMYT).

Main Results:

• Regularized SIs derived from hyper-spectral data demonstrated consistently higher accuracy for predicting grain yield compared to standard SIs.
• The developed method outperformed commonly used vegetation indices for predicting agronomic traits.

Conclusions:

• Regularized SIs provide an effective approach for leveraging routinely generated HTP data in agriculture.
• The methodology is applicable to genetic studies involving high-dimensional phenotypes in various organisms, including humans and model organisms.