Hello everyone,

I am performing batch effect correction on bulk RNA-seq data from human adipose tissue. Samples were collected from two depots, subcutaneous (SAT) and omental (OAT), across subjects with a wide BMI range (lean, overweight, and obese). Sequencing was performed in two separate batches.

I used limma::removeBatchEffect() in R for visualization and PCA, with the voom + lmFit workflow for differential expression. My full design matrix includes: Batch, Depot (SAT/OAT), and BMI status (lean/overweight/obese).

My question is: should Depot be included as a covariate in the batch correction model, or should the correction be kept agnostic of depot?

My concern is that SAT and OAT have genuinely different transcriptomic profiles. If I include Depot in the model, the correction preserves those biological differences while removing batch. If I omit it, batch correction may partially absorb the depot signal, especially if the two depots are not perfectly balanced across batches.

What is the recommended approach, and are there any risks of over-correction if Depot is included?

Thank you in advance!