Accounting for overlapping annotations in genomic prediction models of complex traits

Area of Science:

• Genomics
• Quantitative Genetics
• Bioinformatics

Background:

• Genomic prediction models are widely used in livestock and plant breeding to predict phenotypes from genotyping data.
• Increasingly granular genomic annotations offer insights into functionally important genomic regions.
• The existing BayesRC model, while useful, cannot accommodate markers with multiple annotations.

Purpose of the Study:

• To develop novel Bayesian approaches (BayesRC+ and BayesRC[Formula: see text]) to address multi-annotated markers in genomic prediction.
• To evaluate the performance of these new models on simulated and real-world data, including pig breeding populations.
• To explore strategies for constructing informative annotations from public databases.

Main Methods:

• Proposed two Bayesian models: BayesRC+ (cumulative) and BayesRC[Formula: see text] (preferential) to model contributions from multiple annotation categories.
• Tested models on simulated data with diverse genetic architectures and annotation types.
• Applied models to pig backcross population data using annotations from PigQTLdb.

Main Results:

• Both BayesRC+ and BayesRC[Formula: see text] demonstrated modest improvements in prediction quality with informative annotations on simulated and real data.
• BayesRC+ effectively prioritized multi-annotated markers based on posterior variance.
• BayesRC[Formula: see text] provided valuable interpretations of informative annotations for multi-annotated markers.

Conclusions:

• BayesRC+ and BayesRC[Formula: see text] enhance prediction accuracy and marker prioritization when utilizing trait-relevant annotations.
• These models offer significant biological insights into the genetic architecture of complex traits.
• Careful construction of annotations from public databases is crucial for model performance.