Reciprocal BLUP: A Predictability-Guided Multi-Omics Framework for Plant Phenotype Prediction.

Area of Science:

• Plant genomics and breeding
• Multi-omics data integration
• Systems biology in agriculture

Background:

• Improving crop performance necessitates linking genetic variation to observable traits via molecular pathways.
• Current methods often struggle to integrate diverse omics data for comprehensive phenotype prediction.
• Understanding genotype-environment-microbiome interactions is crucial for sustainable agriculture.

Purpose of the Study:

• To introduce Reciprocal Best Linear Unbiased Prediction (Reciprocal BLUP), a framework for multi-omics integration.
• To quantify cross-layer relationships between genome, metabolome, and microbiome for enhanced phenotype prediction.
• To evaluate the utility of predictability-guided feature selection in plant breeding applications.

Main Methods:

• Developed and applied the Reciprocal BLUP framework to soybean accessions under varying water conditions.
• Evaluated four direction-specific prediction models to assess cross-omics feature predictability.
• Compared phenotype prediction accuracy using predictability-selected features versus random selection and full-feature models.

Main Results:

• Identified an asymmetric relationship where metabolome is predictable from microbiome, but not vice-versa.
• Predictability-guided feature selection significantly improved phenotype prediction accuracy compared to random selection.
• Metabolomic and microbiomic models (MetBLUP, MicroBLUP) outperformed genomic prediction (GBLUP) for biomass traits under drought.

Conclusions:

• The metabolome acts as a key mediator linking genotype, environment, and microbiome composition in plants.
• Reciprocal BLUP provides a biologically interpretable method for integrating multi-omics data in plant breeding.
• This framework enhances phenotype prediction and guides omics-based feature selection for crop improvement.