SCZ_MWAS_meta_analysis_GSE80417_GSE152027_GSE84727.csv

Objective: This study sought to determine if the R package LDpred2, designed for polygenic risk score creation for genome-wide association studies using summary statistics, could be adapted for deriving DNA methylation scores from methylome-wide association studies. Recognizing that linkage disequilibrium, used as prior in LDpred2, does not apply to methylation, we explored co-methylated regions and topologically associating domains as alternative structural priors for correlation between methylation sites. A genomic sliding-window approach was also tested. The performance of the LDpred2-based models was evaluated on methylation data from schizophrenia and control samples (N=1,227). Results: LDpred2 models employing topologically associating domains and sliding window clusters as priors performed similarly to existing methods, explaining approximately 3.6% of schizophrenia phenotypic variance. The co-methylated regions model underperformed due to insufficient clustering of probes. The similarity in performance between the model using topologically associating domains and a null model consisting of random clusters suggests that the structural information provided by these domains enhances performance only marginally. In conclusion, while LDpred2 can be adapted for methylation data, it does not substantially enhance methylation score performance over existing methods, and the choice of structural prior may not be a critical factor .