Development of Optimized Phenomic Predictors for Efficient Plant Breeding Decisions Using Phenomic-Assisted Selection...

Area of Science:

• Plant breeding
• Agricultural science
• Machine learning in agriculture

Background:

• Genomic-assisted breeding is advanced, but phenomic-assisted methods lag in cultivar development.
• High-dimensional phenotyping data offers opportunities to improve breeding program efficiency.

Purpose of the Study:

• To optimize breeding program operational efficiencies using phenomic data.
• To develop a prescriptive sensor package for high-throughput phenotyping.
• To identify optimal phenotypic traits for predicting crop yield.

Main Methods:

• Collected phenomic and seed yield data for 292 soybean accessions across six environments.
• Applied Random Forest (ML algorithm) to map relationships between phenomic traits and seed yield.
• Utilized feature importance and a Genetic Algorithm (GA) to select optimal phenotypic traits (wavebands).

Main Results:

• Identified a suite of in-season phenomic traits predictive of seed yield.
• Demonstrated the fusion of ML and optimization techniques for trait selection.
• Showcased the potential to decrease dependence on resource-intensive end-season phenotyping.

Conclusions:

• Phenomic prediction fused with ML and optimization offers a template for crop breeding.
• This approach can be adapted for any crop species and breeding objective.
• Enables more efficient breeding programs through early trait selection.