UC Berkeley

Confounding by cellular heterogeneity has become a major concern for epigenome-wide association studies (EWAS) in peripheral blood samples from population and clinical studies. Adjusting for white blood cell percentage estimates produced by the minfi implementation of the Houseman algorithm (minfi) during statistical analysis is now an established method to account for this bias in adults. However, minfi has not been benchmarked against white blood cell counts in children that may differ substantially from the reference dataset used in its estimation. We compared estimates of white blood cell type percentages produced by two methods, minfi and differential cell count (DCC), in a birth cohort at two time points (birth and 12 years of age). We found that both minfi and DCC had similar trends as children aged, and neither count method differed by sex among newborns (P > 0.10). However, minfi estimates did not correlate well with DCC in samples from newborns (ρ = -0.05 for granulocytes; ρ = -0.03 for lymphocytes). In older children, correlation improved substantially (ρ = 0.77 for granulocytes; ρ = 0.75 for lymphocytes), likely due to increasing similarity with minfi's adult reference data as children aged. Our findings suggest that the minfi method may provide suitable estimates of white blood cell composition for samples from adults and older children, but may not currently be appropriate for EWAS involving newborns or young children.