Integrating gene expression data via weighted multiple kernel ridge regression improved accuracy of genomic...

Area of Science:

• Animal Breeding and Genetics
• Bioinformatics
• Genomics

Background:

• Gene expression profiles offer valuable insights for predicting breeding values and phenotypes.
• Practical breeding programs often lack transcriptomic data, relying solely on genomic data.
• Predicting gene expression from genetic markers presents a solution for integrating transcriptomic information.

Purpose of the Study:

• To develop and evaluate a novel method, weighted multiple kernel ridge regression (WMKRR), for integrating genomic and genetically predicted transcriptomic data.
• To compare the predictive ability of WMKRR against traditional genomic best linear unbiased prediction (GBLUP) and combined genomic and transcriptomic best linear unbiased prediction (GTBLUP).

Main Methods:

• Extended kernel ridge regression (KRR) to weighted multiple kernel ridge regression (WMKRR).
• Utilized a multiple kernel learning (MKL) approach to integrate genomic data and transcriptomic data predicted from genetic markers.
• Evaluated WMKRR using simulated data based on the CattleGTEx dataset and real dairy cattle data, employing both feature selection and non-feature selection scenarios.

Main Results:

• WMKRR demonstrated superior predictive abilities compared to GBLUP and GTBLUP in both simulated and real dairy cattle datasets.
• WMKRR achieved average improvements in predictive ability of 1.12%-3.23% over GBLUP and GTBLUP in simulated data.
• In real dairy cattle data, WMKRR showed average improvements of 5.56%-8.41% over GBLUP and GTBLUP across different validation scenarios.

Conclusions:

• The WMKRR model effectively integrates genomic and genetically predicted transcriptomic data, outperforming traditional genomic prediction models.
• This study highlights the potential for enhancing genomic breeding applications by leveraging omics data without increased sequencing expenses.