Cell-type specific analysis of heterogeneous methylation signal using a Bayesian model-based approach [article]

Daniel W Kennedy, Nicole M White, Miles C Benton, Rodney A Lea, Kerrie Mengersen
2019 bioRxiv   pre-print
Epigenome-wide association studies (EWASs) are often performed using DNA collected from heterogeneous cell samples yet there is often good reason to expect methylation variation in specific cell subtypes to be associated with disease. While statistical methods have been established to ascertaining cell-type fractions and to account for cell-type heterogeneity, methods to identify cell-type specific signals from heterogeneous samples in EWASs are lacking. We present a Bayesian model-based
more » ... h for inferring cell-type specific differential methylation (Bayes-CDM) from heterogeneous blood samples. The method uses a logit-normal sampling distribution and incorporates a priori knowledge of cell-type lineage. The method is tested in a case-control EWAS design by using mixed blood cell methylation data to estimate cell-type specific effects on sex as a binary outcome, whereby blood cell subtype methylation data was available as the ground-truth reference. Results: Cell-type specific differential methylation loci selected using Bayes-CDM contained most (> 84%) of the ground-truth loci for all cell-types. Estimated cell-type specific effect sizes exhibited a very high degree of correlation with the ground-truth values even after sex-chromosome CpG loci were removed (R-sq > 0.90). These findings provide compelling evidence that Bayes-CDM can add value to mixed cell EWASs by facilitating the detection of phenotype associations at the cell-type level.
doi:10.1101/682070 fatcat:xizn7fquijeefcpun5i2x4rttm