Incorporating kernelized multi-omics data improves the accuracy of genomic prediction

Area of Science:

• Animal breeding and genetics
• Genomics and bioinformatics
• Quantitative genetics

Background:

• Genomic selection (GS) has transformed breeding, with growing interest in multi-omics data (genome, transcriptome, metabolome).
• Integrating multi-omics data for GS faces challenges in model building and limited specimen availability.
• Previous research has primarily focused on genomic data for GS.

Purpose of the Study:

• To investigate the potential of integrating transcriptome data into genomic selection models.
• To develop and evaluate novel multi-omics prediction models for improved accuracy.
• To assess the performance of different models in a Chinese Simmental beef cattle population.

Main Methods:

• Utilized a Cosine kernel to create genomic (G) and transcriptomic (T) matrices.
• Developed five kernel-based prediction models: GBLUP, TBLUP, MBLUP, mssBLUP, and wmssBLUP.
• Integrated genotyped and transcribed individuals using BLUP for GS.

Main Results:

• Multi-omics BLUP (MBLUP) significantly outperformed genomic BLUP (GBLUP).
• Weighted multi-omics single-step BLUP (wmssBLUP) and mssBLUP showed average accuracy improvements of 4.18% and 3.37% over GBLUP, respectively.
• wmssBLUP accuracy increased with a higher proportion of transcribed cattle in the resource population.

Conclusions:

• Inclusion of transcriptome data in GS can substantially improve prediction accuracy.
• The weighted multi-omics single-step BLUP (wmssBLUP) model is a promising approach for current breeding scenarios with abundant genotyped individuals and fewer transcribed ones.