Incorporating genomic and transcriptomic effects in joint linear and non-linear structural models for predicting...

Area of Science:

• Animal Genomics and Breeding
• Transcriptomics and Gene Expression Analysis
• Quantitative Genetics and Trait Prediction

Background:

• Livestock phenotypes result from complex interactions between genetic variation and regulatory mechanisms.
• Predicting complex traits is crucial for efficient animal breeding but remains challenging.
• Transcriptomic data offer insights into regulatory signals bridging genotypes and phenotypes.

Purpose of the Study:

• To assess the predictive value of blood transcriptomic data for six immune, stress, and production traits in Duroc pigs.
• To compare the performance of transcriptomic data alone versus combined with genomic information.
• To evaluate different statistical models and feature selection methods for trait prediction.

Main Methods:

• Utilized blood transcriptomic and genomic (SNP) data from 255 Duroc pigs.
• Employed Bayesian regression (BayesC, RKHS) and neural network linear mixed models.
• Applied Partial Least Squares (PLS) for transcriptomic feature selection.

Main Results:

• High prediction accuracies achieved for immunity-related traits (e.g., gamma delta T cells, leukocyte counts) using transcriptomic data (r=0.74, r=0.67).
• Moderate improvement for cortisol prediction (r=0.39); SNP-based models excelled for carcass weight (r=0.45).
• PLS feature selection identified key genes (MAF, SOX13, DDIT4, FOS) and improved prediction efficiency.

Conclusions:

• Blood transcriptomics substantially enhance prediction for traits relevant to the sampled tissue.
• SNP-based models are superior for traits less biologically related to blood.
• Feature selection is critical for optimizing prediction performance, computational efficiency, and gene discovery.