Positional effects revealed in Illumina methylation array and the impact on analysis

We aimed to prove the existence of positional effects in the Illumina methylation beadchip data and to find an optimal correction method.

Three HumanMethylation450, three HumanMethylation27 datasets and two EPIC datasets were analyzed. ComBat, linear regression, functional normalization and single-sample Noob were used for minimizing positional effects. The corrected results were evaluated by four methods.

We detected 52,988 CpG loci significantly associated with sample positions, 112 remained after ComBat correction in the primary dataset. The pre- and postcorrection comparisons indicate the positional effects could alter the measured methylation values and downstream analysis results.

Positional effects exist in the Illumina methylation array and may bias the analyses. Using ComBat to correct positional effects is recommended.