A data-adaptive Bayesian regression approach for polygenic risk prediction.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Polygenic risk scores (PRS) are crucial for genetic risk prediction.
• Existing methods for PRS construction have limitations in accommodating diverse genetic architectures.

Purpose of the Study:

• Introduce NeuPred, a unified Bayesian regression framework for PRS construction.
• Improve prediction accuracy for complex diseases by allowing flexible prior choices.
• Develop a strategy for automatic selection of optimal chromosome-level priors.

Main Methods:

• Developed a unified Bayesian regression framework named NeuPred.
• Proposed a summary-statistics-based cross-validation strategy for selecting chromosome-level priors.
• Evaluated NeuPred using simulation studies and real-world disease datasets.

Main Results:

• NeuPred demonstrated substantial and consistent improvements in predictive r2 over existing methods.
• Observed significant variability in prior preference across different chromosomes for the same disease.
• NeuPred showed comparable or superior computational efficiency to state-of-the-art Bayesian methods.

Conclusions:

• NeuPred offers an advanced framework for PRS construction, enhancing genetic risk prediction.
• The proposed prior selection strategy optimizes PRS accuracy by accounting for chromosome-specific genetic architectures.
• NeuPred provides a computationally efficient and accurate tool for complex disease genetic risk assessment.