Towards holistic phenotype prediction beyond genotypic data.

Area of Science:

• Plant and Animal Breeding
• Genetics
• Bioinformatics

Background:

• Genomic selection (GS) revolutionizes breeding by predicting phenotypes from genetic data.
• Current GS models explain only a fraction of observed phenotypic variation.
• There is a need to integrate diverse data types for enhanced prediction accuracy.

Purpose of the Study:

• To review and categorize strategies for integrating non-genomic data into genomic selection.
• To explore methods that enhance phenotypic prediction beyond genetic information.
• To provide a comprehensive overview of multi-data integration in breeding.

Main Methods:

• Categorization of data integration strategies into five types: eliminate, facilitate, aggregate, incorporate, and modulate.
• Review of methods leveraging environmental, phenotypic, and other biological data.
• Discussion of advanced modeling techniques, including deep learning (e.g., CNNs).

Main Results:

• Five distinct data integration strategies offer varying benefits for phenotypic prediction.
• Facilitating, aggregating, incorporating, and modulating methods show promise for improving GS accuracy.
• Explicitly modeling interactions and transforming data for advanced models are key approaches.

Conclusions:

• Multi-data phenotypic prediction offers a holistic approach to understanding complex biological systems.
• Integrating diverse data types significantly enhances prediction accuracy in breeding programs.
• Future research should focus on developing comprehensive prediction models combining genomics and other data sources.