Improving Genomic Prediction for Feed Efficiency Traits in Beef Cattle Using Microbiome as an Intermediate Omics

Authors

• Egodage Bhagya Sewwandi Samarakoon, University of Nebraska-LincolnFollow

First Advisor

Tianjing Zhao

Committee Members

Matthew L. Spangler, Réka Howard

Date of this Version

5-2026

Document Type

Thesis

Citation

A thesis presented to the faculty of the Graduate College at the University of Nebraska in partial fulfillment of requirements for the degree of Master of Science

Major: Animal Science

Under the supervision of Professor Tianjing Zhao

Lincoln, Nebraska, May 2026

Comments

Copyright 2026, Egodage Bhagya Sewwandi Samarakoon. Used by permission

Abstract

Feed efficiency traits, including feed conversion ratio (FCR), average daily gain (ADG), and average daily dry matter intake (ADDMI), are economically important in beef cattle production. Emerging evidence suggests that host genetics influences feed efficiency traits through the microbiome. This study evaluates the integration of microbiome data as an intermediate layer in the Neural Network Mixed Model (NNMM), which hierarchically models genotype–microbiome–phenotype relationships as a unified regulatory network to improve genomic prediction accuracy. To reduce the computational burden of NNMM, microbiome features dimension were reduced using principal component analysis (PCA) and direct feature selection of top-ranked open reading frames (ORFs) via microbiome-wide association studies (MWAS). Two baseline models were used to compare the prediction accuracy of NNMM: (1) Genomic Best Linear Unbiased Prediction (GBLUP) and (2) a two- kernel genomic–microbiome model (G+M) incorporating the microbiome as an additional non-genetic random effect in a linear mixed model. Additionally, we evaluated different microbiome relationship matrix (MRM) construction methods. A beef cattle population (n = 722) with complete phenotypic (FCR, ADG and ADDMI), genomic (922,208 SNPs) and rumen microbiome (16,583 ORFs) records were analyzed.

Results demonstrated that microbiability estimates are sensitive to both the chosen MRM structure and dimension-reduction method. NNMM consistently outperformed the other approaches. In detail, for the low-heritability traits ADG (h² = 0.19) and FCR (h² = 0.13), significant improvements were observed when trait relevant ORFs were used as intermediate omics in NNMM: the NNMM model improved prediction accuracy for ADG from 0.08 (GBLUP) to 0.28 and for FCR from 0.08 (G+M) to 0.26. For the high-heritability trait ADDMI (h² = 0.44), the best NNMM model increased prediction accuracy from 0.30 (GBLUP and G+M) to 0.32 when using PCs of microbiome data as intermediate omics that captured 100% of the microbiome variance. Our results demonstrate the practical advantage of the NNMM model in improving the genomic prediction accuracy of feed efficiency in beef cattle by incorporating the microbiome as an intermediate omics, establishing this approach as a microbiome-enabled genomic selection strategy for sustainable national beef production.

Advisor: Tianjing Zhao