Blood-based epigenome-wide analyses of chronic low-grade inflammation across diverse population cohorts - Bayesian EWAS and CRP Predictors

This dataset contains four files that are separated into two groups based on the analyses they inform in the manuscript by Hillary et al. entitled ‘Blood-based epigenome-wide analyses of chronic low-grade inflammation across diverse population cohorts’. The manuscript is available as: https://doi.org/10.1101/2023.11.02.23298000.

The first group contains one file called ‘crp_bayespr_basic_model_hillaryetal.csv’. This file contains the output of a Bayesian penalised regression model that was used to perform an epigenome-wide association study on blood C-reactive protein or CRP levels in Generation Scotland participants (N=17,936). The model examined the association between 752,722 CpG sites and log-transformed blood CRP levels. CRP levels were adjusted for age and sex prior to entry into the models. CpG beta-values were regressed on age, sex, estimated white blood cell proportions and experimental batch. Residuals were scaled to mean zero and unit variance and entered into the Bayesian model. The coefficients reflect these standardised variables. Results from a standard linear epigenome-wide association study are available on the EWAS Catalog.

The second group contains three files called ‘elastic_net_predictor_for_crp.csv’, ‘bayespr_predictor_for_crp.csv’ and ‘pca_elnet_prediction_object_for_crp.rds’. These files contain CpG sites and their weights from three different prediction methods. The methods were elastic net regression, Bayesian penalised regression and elastic net combined with principal component analysis, respectively. The predictors for C-reactive protein levels were trained in Generation Scotland and can be applied to external cohorts. Prediction scripts and the predictors (grouped together) are available in the following repository: https://doi.org/10.5281/zenodo.10154736.