On the question of correct use of replicates in quantitative label-free proteomics

Label-free quantitation is the most popular method in proteomics for assessing changes in protein concentrations. However, practical aspects like the optimal use of technical replicates, the impact of removing low-identified protein runs, and the effect of combining information from technical replicates for subsequent differential expression analysis remain debated. This study utilized five LFQ workflows: MaxQuant + Perseus, FragPipe + MSstats, Proteome Discoverer, DirectMS1Quant, and IdentiPy + IQMMA. Previously published data sets acquired for three-species proteomes using Orbitrap FTMS consisted of spikes of Escherichia coli, yeast, and human lysates with known concentration changes that were used for benchmarking the workflows. All tested workflows gave fairly similar results in terms of the number of differentially expressed proteins (DEPs) and quantitative false discovery rate (FDR). Adding more technical replicates either increased the number of DEPs or decreased the FDR, depending on the workflow. Eliminating runs with the lowest number of protein identifications led to an increase in the number of DEPs, but at the cost of elevated FDR, thus reducing the accuracy and precision of protein fold change estimations. The Match-Between-Runs option provides additional DEPs and does not increase empirical FDR in most methods. We found that the selected set of proteomics workflows turned out to be different in answering the practical questions raised above, even for the simple artificial benchmark data set. Our results should serve as a starting point and encourage researchers to more thoroughly test their own approaches in real-world problems.