Extend mixed models to multilayer neural networks for genomic prediction including intermediate omics data

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Increasing availability of diverse intermediate omics data (DNA methylation, gene expression, protein abundance) necessitates advanced genomic evaluation methods.
• Omics data reveal multilayer regulatory networks from genotypes to phenotypes, requiring integrated analytical approaches.

Purpose of the Study:

• To develop a novel method, NN-MM, for incorporating intermediate omics data into genomic evaluation.
• To model the multilayer regulatory pathways from genotypes through intermediate omics features to phenotypes.

Main Methods:

• Extended conventional linear mixed models (MM) to multilayer artificial neural networks (NN), creating the NN-MM framework.
• Utilized linear mixed models for marker effects/genetic values on omics features and neural network activation functions for nonlinear genotype-phenotype relationships.
• Implemented NN-MM in the open-source JWAS package, designed to handle missing omics data.

Main Results:

• NN-MM demonstrated significantly improved prediction performance compared to existing single-step methods for genomic prediction using intermediate omics data.
• The method effectively models nonlinear relationships between intermediate omics features and phenotypes.
• NN-MM shows robustness in handling various missing data patterns in omics measures.

Conclusions:

• NN-MM provides a powerful and flexible framework for integrating multilayer omics data in genomic evaluation.
• The method offers superior prediction accuracy and better handling of data complexities than previous approaches.
• The open-source JWAS package facilitates the application of NN-MM in biological and agricultural research.