Adding gene transcripts into genomic prediction improves accuracy and reveals sampling time dependence.

Area of Science:

• Genomics
• Animal Breeding
• Bioinformatics

Background:

• High-quality 'omics' data, including transcriptomics, can enhance genomic prediction accuracy.
• Genomic prediction is crucial for improving phenotypes and genetic merit in animals and plants.

Purpose of the Study:

• To assess parametric and nonparametric models for genomic prediction using transcriptomics data.
• To evaluate new models (GTCBLUP, GTBLUP) for their ability to leverage multi-omics data.
• To compare the predictive performance of different models across 13 complex traits in mice.

Main Methods:

• Implemented parametric models using Best Linear Unbiased Prediction (BLUP).
• Utilized Gradient Boosting Machine (GBM) for nonparametric models.
• Developed and tested GTCBLUP and GTBLUP models, analyzing covariance between omics layers.

Main Results:

• GBM models captured more phenotypic variation, but BLUP models showed comparable predictive performance for most traits.
• Transcriptomics data improved prediction accuracy, especially when measured close to phenotypic data collection.
• The GTCBLUP model demonstrated the highest accuracy for breeding values in 9 out of 13 traits.

Conclusions:

• GTBLUP is recommended for phenotype prediction, while GTCBLUP is suggested for breeding value prediction.
• Combining omics layers did not consistently outperform single-omics models for phenotype prediction.
• Transcriptomics data significantly contributes to predicting complex traits, particularly when time-matched with phenotypic measurements.