I am working on preprocessing an Illumina DNA methylation dataset for downstream DMP analysis using limma. My EWAS model will include the intervention/exposure of interest as well as cell-type proportions as covariates.

I plan to estimate cell-type proportions using a reference-free approach (e.g., RefFreeEWAS or TOAST). My current preprocessing pipeline includes QC and normalization, probe filtering (e.g., cross-reactive probes, SNP-affected probes, sex chromosomes), filtering probes with high missingness, batch correction using ComBat, and removal of outlier probes.

My question is specifically about where cell-type proportion estimation should occur in this pipeline.

For reference-free methods such as RefFreeEWAS/TOAST, is it appropriate to estimate cell-type proportions after batch correction (e.g., on ComBat-adjusted beta values), or is it recommended to estimate them on an intermediate matrix prior to batch correction, and then include those estimated proportions as covariates in the downstream limma model?

I would appreciate any guidance or references on best practice for ordering these steps.