Effective genetic-risk prediction using mixed models.

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Predicting genetic risk for complex diseases is challenging due to intricate disease phenotypes and case-control sampling.
• Current methods often struggle to effectively integrate information from single nucleotide polymorphisms (SNPs) and whole-genome similarities.

Purpose of the Study:

• To introduce Genetic Risk Scores Inference (GeRSI), a new statistical approach for enhanced genetic risk prediction.
• To evaluate GeRSI's performance against state-of-the-art methods using simulations and real-world case-control data.

Main Methods:

• GeRSI combines fixed-effects models (aggregating SNP effects) and random-effects models (using genome-wide similarities).
• The approach is integrated within the liability-threshold model framework.
• Performance was assessed using extensive simulations and applied to seven phenotypes from the Wellcome Trust Case Control Consortium (WTCCC).

Main Results:

• GeRSI consistently outperformed current state-of-the-art approaches in simulations.
• For hypertension (HT), GeRSI improved the area under the ROC curve (AUC) from 54% to 59%.
• For bipolar disorder (BD), GeRSI increased AUC from 55% to 62%, substantially elevating relative risk for top-predicted individuals.

Conclusions:

• GeRSI offers superior genetic risk prediction, particularly for highly polygenic diseases.
• The integration of random effects in GeRSI is most beneficial for complex conditions like hypertension and bipolar disorder.
• GeRSI demonstrates significant potential for improving clinical risk assessment and stratification.