POISE: Spectral Inference of Parent-of-Origin Effects in Unlabeled Genomic Data.

Area of Science:

• Genetics
• Bioinformatics
• Machine Learning

Background:

• Parent of Origin Effects (POEs) influence traits like growth, metabolism, and neurodevelopment.
• Traditional POE detection requires costly family-based studies.
• There is a need for methods that infer POEs without inheritance data, complementing Genome-Wide Association Studies (GWAS).

Purpose of the Study:

• To develop a novel computational method for inferring Parent of Origin Effects (POEs).
• To create a method that does not require family data, making POE analysis more accessible.
• To identify new genetic variants associated with complex traits through POE analysis.

Main Methods:

• Developed Parent of Origin Inference via Spectral Estimation (POISE), a machine learning-based method.
• Utilized spectral decomposition for POE inference and non-parametric bootstrap for confidence intervals.
• Applied POISE to UK Biobank GWAS data for BMI, LDL, and HDL cholesterol.

Main Results:

• POISE demonstrated robustness and improved accuracy over existing methods in simulations.
• The method identified 134 novel variants associated with POEs in lipid metabolism, immune regulation, and growth.
• POISE successfully recovered established POE loci in the UK Biobank cohort.

Conclusions:

• POISE is a powerful, robust, and scalable method for inferring Parent of Origin Effects from GWAS data.
• The method facilitates the discovery of novel genetic associations without the need for family data.
• POISE enhances our understanding of genetic architecture in complex human traits.