Accounting for longitudinal peak quality metrics with MSstats+ enhances differential analysis in proteomic experiments with data-independent acquisition

Abstract

Mass spectrometry-based proteomics with data-independent acquisition benefits from advanced instrumentation and computational analysis. Despite continued improvements, the quality of quantification may be poor for some measurements. As the scale of proteomic experiments increases, these poor-quality measurements are challenging to characterize by hand, yet they undermine the detection of differentially abundant proteins and the downstream biological conclusions. We introduce MSstats+, a computational workflow that takes as input not only peak intensities reported by tools such as Spectronaut, but also quality metrics such as peak shape and retention time, and longitudinal run order profiles of these metrics. MSstats+ translates these metrics into a single measure of quality, and downweights poor quality measurements when detecting differentially abundant proteins. The method offers a natural treatment of missing values, weighting the imputed values according to the quality metrics in the run. We demonstrate the accuracy of the resulting differential analysis in four experiments: two custom benchmarking studies with intentionally induced anomalies, a controlled mixture of proteomes, and a large-scale clinical investigation. MSstats+ is implemented in the family of open-source R/Bioconductor packages MSstats.