Set effects are usually quite imponent and mask sample characteristics when dealing with human samples and TMT-labeled non-targeted proteomics.
What is in your view the best approach to preserve the experimental differences while flattening down set effects (technical artifacts)?