A polygenic score method boosted by non-additive models.

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Dominance heritability is increasingly recognized in complex traits.
• Current polygenic score (PGS) methods often neglect non-additive genetic effects.

Purpose of the Study:

• Introduce GenoBoost, a novel PGS framework.
• Incorporate additive and non-additive genetic effects, focusing on genetic dominance.
• Enhance predictive accuracy and biological insights from PGS.

Main Methods:

• Statistical boosting theory for optimal score derivation.
• Efficient implementation for large-scale cohort analysis.
• Benchmarking against seven common PGS methods using UK Biobank data.

Main Results:

• GenoBoost demonstrates competitive predictive performance, outperforming other methods for several traits.
• Improved prediction for autoimmune diseases by including non-additive effects in the MHC locus.
• Identified non-zero genetic dominance effects for numerous variants, improving psoriasis prediction by 2.5%.

Conclusions:

• GenoBoost offers enhanced accuracy and biological insights by incorporating non-additive genetic effects.
• The framework can infer modes of genetic inheritance without prior knowledge.
• GenoBoost prioritizes genetic loci with previously unreported genetic dominance.