Extended Data Fig. 2: Benchmarking computational analysis of individual CF–MS datasets.

a, Measures of association used to quantify the similarity of two protein chromatograms in published CF–MS studies. Bottom row indicates the incorporation of external genomic datasets⁷⁷. b, Ranks of each measure of association in identifying protein pairs in the same protein complex, left, or annotated to the same GO term, right, across individual CF–MS datasets. c, Number of peaks detected in 20 CF–MS datasets by fitting a mixture of Gaussians to each protein chromatogram. d, Recovery of known protein complexes in the 20 CF–MS datasets from c, scoring only chromatograms that could be fit with a mixture of Gaussians (r² ≥ 0.5) and comparing the 24 different measures of association shown in Fig. 2 with the co-apex score. Inset text shows the median AUC for each measure of association. e, As in d, but for proteins annotated to the same GO term. f, Recovery of known protein complexes, top, and proportion of originally quantified proteins, bottom, when filtering profiles not detected in some minimum number of fractions, using mutual information as a measure of profile similarity. g, Mean number of protein groups identified, top, and recovery of proteins annotated to the same GO term, bottom, for three approaches to label-free quantification implemented in MaxQuant.