Using LDpred2 to adapt polygenic risk score techniques for methylation score creation.

Area of Science:

• Genetics
• Epigenetics
• Computational Biology

Background:

• Polygenic risk scores (PRS) are crucial for predicting disease risk using genome-wide association studies (GWAS).
• The LDpred2 package is a popular tool for PRS creation from GWAS summary statistics, utilizing linkage disequilibrium (LD) as a prior.
• LD, a measure of non-random association between alleles, is not directly applicable to DNA methylation data.

Purpose of the Study:

• To evaluate the adaptability of the LDpred2 package for deriving DNA methylation scores from methylome-wide association studies (MWAS).
• To explore alternative structural priors, such as co-methylated regions and topologically associating domains (TADs), to model correlations between methylation sites.
• To assess the performance of LDpred2-based models using methylation data from schizophrenia and control samples.

Main Methods:

• Adaptation of the LDpred2 R package for MWAS data.
• Implementation of alternative structural priors: co-methylated regions, TADs, and genomic sliding windows.
• Performance evaluation using schizophrenia methylation data (N=1,227) and comparison with existing methods.

Main Results:

• LDpred2 models using TADs and sliding window clusters as priors showed performance comparable to existing methods, explaining ~3.6% of schizophrenia phenotypic variance.
• A model using co-methylated regions underperformed due to inadequate probe clustering.
• The marginal performance gain from TADs suggests limited benefit from this specific structural prior.

Conclusions:

• The LDpred2 package can be adapted for DNA methylation score derivation.
• Current adaptation does not significantly outperform existing methods for methylation analysis.
• The choice of structural prior (e.g., TADs) may not be a critical determinant of performance in this context.