Multi-omics analysis

Multi-Omics Data Integration and Transcriptome Imputation for Improved Genetic Evaluation in Hanwoo cattle

---

Abstract

This study aimed to evaluate the utility of transcriptome imputation for livestock breeding by constructing a gene expression prediction database and applying mixed-model frameworks that incorporate correlations between genomic and transcriptomic effects. Integrating multi-omics information is essential for understanding complex biological mechanisms, and recent advances in high-throughput technologies have greatly expanded transcriptomic resources. However, linear mixed models widely used in livestock breeding assume independence among random effects, limiting their ability to capture regulatory interdependencies among SNPs, gene expression, and epigenetic factors. In this study, a reference dataset consisting of genotypes, gene expression levels, and phenotypes from 123 animals was used to train an elastic-net-based imputation model analogous to the PredictDB-PrediXcan framework. SNPs served as predictors, and gene expression levels were used as response variables. The resulting weights were applied to SNP genotypes from 17,000 animals generated through the BioGreen project. SNP imputation and genomic-coordinate liftover were performed to harmonize datasets prior to analysis. Using observed phenotypes, imputed gene-expression levels, and SNP data, four models were compared: (1) G0, a standard GBLUP model; (2) T, a model fitting genetically regulated expression; (3) G0-T_GREML, which fits genomic and expression effects simultaneously without estimating their covariance; and (4) G0-T_COREGREML, which additionally estimates covariance between random effects. The average imputation accuracy for gene expression was 0.55, and the heritability of imputed transcriptomic expression was consistently estimated at 0.01-0.02 under both the G0-T_GREML and G0-T_COREGREML models. These results indicate that the transcriptomic component captures a small but detectable proportion of genetic variance. Likelihood ratio tests comparing the G0 and G0-T_GREML models revealed significant differences across all traits (P = 1.06E-04-3.45E-02), demonstrating that imputed transcriptome information explains a meaningful fraction of phenotypic variation. The null hypothesis σ²t/σ²y = 0 was also rejected for all traits (P = 7.60E-04-4.90E-02), confirming that the transcriptomic variance component, although limited in magnitude, is statistically significant. Importantly, the ratio of genomic variance (σ²g/σ²y) remained nearly identical between the G0 and G0-T_GREML models (e.g., 0.41 vs. 0.39 for CW), indicating no evidence of overfitting despite the inclusion of an additional kernel. Variance component estimates in the G0-T_GREML and G0-T_COREGREML models showed stable patterns. The genomic proportion of variance (σ²g/σ²y) ranged from 0.34 to 0.47, while the transcriptomic proportion (σ²t/σ²y) remained consistently between 0.01 and 0.02 across all traits. Prediction accuracy results were similarly aligned. Genomic prediction accuracy in the G0 model ranged from 0.689 to 0.724, and σ²g-based accuracies in the G0-T_GREML and G0-T_COREGREML models showed comparable ranges of 0.677-0.718. In contrast, prediction accuracy for the transcriptomic component was lower (0.266-0.407), likely reflecting the limited sample size of the reference transcriptome panel and reduced statistical power for estimating expression-mediated effects.

Keywords: 2026

How to Cite:

Choi, I., Chung, K., Chung, Y., Dinh, P., Kim, J., Ko, S., Lee, S. & Oh, H., (2026) “Multi-Omics Data Integration and Transcriptome Imputation for Improved Genetic Evaluation in Hanwoo cattle”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286539. doi: https://doi.org/10.31274/wcgalp.24042

Rights: 1