Multimodality, interaction modeling, and multimodule architectures in genomic prediction: A unified conceptual...

Area of Science:

• Plant breeding and quantitative genetics
• Statistical learning and data science

Background:

• Genomic, environmental, and phenomic data are rapidly expanding, transforming plant breeding.
• Inconsistent use of terms like multimodal, interaction modeling, and multimodule architecture creates ambiguity.

Purpose of the Study:

• To clarify terminology and model design principles in genomic prediction.
• To improve methodological transparency and facilitate fair comparison of prediction approaches.
• To strengthen communication among quantitative geneticists, data scientists, and breeders.

Main Methods:

• Proposed a framework to distinguish between multimodality, interaction modeling, and multimodularity.
• Used conceptual and literature-based examples from wheat breeding for illustration.
• Framed approaches as statistical learning methods for modeling genotype-environment-phenotype relationships.

Main Results:

• Demonstrated that multimodality, interaction modeling, and multimodularity are conceptually independent.
• Illustrated how these dimensions can be independently implemented or combined.
• Emphasized interpretation of model design principles over new experimental results.

Conclusions:

• Clear terminology is crucial for advancing genomic prediction in plant breeding.
• The proposed framework enhances understanding of complex data integration and modeling strategies.
• Accurate modeling of genotype-environment-phenotype relationships is key for effective breeding programs.