Computational Pipeline

The database of complexes was manually compiled and curated from COMPLEAT and CORUM by Ori et al., 2016, and quantified proteins from all published datasets considered for the analysis were mapped accordingly. A subset of manually curated protein complexes were classified as ‘well-defined’ (see Ori et al., 2016). For further analysis only protein complexes with at least 5 quantified members were considered in each dataset, respectively. As a general principle, we used the median co-abundance of proteins within a complex as a proxy to differentiate between stable and variable complexes. To compare the extent of complex stability and variability, correlations were ranked within each dataset; finally the median rank of each complex as recovered from each considered dataset was calculated, and complexes were sorted accordingly. The top quantile (25%) of these complexes were considered to be highly stable (Pearson’s r > 0.46), whereas the lowest quantile were considered highly variable (Pearson’s r < 0.2).

To assess the consistency of the complex variability landscape, we calculated the Spearman correlation of the ranked median co-abundance across datasets (as illustrated in Figure 3). As a reference distribution we permuted the dataset 1000 times, and computed Spearman correlation coefficients across datasets each time. In a two-sided t-test we then compared the real distribution of correlation values with the ones derived from the random permutations of the dataset. This testing set-up does not presume any directionality in the hypothesis testing (two-sided) and is justified due to the normality of the reference distribution.