Supplementary Figure 3: Assessing quality and benchmarking cell envelope protein-protein interactions (cePPIs) in the network.

(a) Proportion of PPIs from each bait CEP filtered at different peptide identification probability score cut-off from SEQUEST/STATQUEST algorithm is plotted against the known interactions curated in EcoCyc. (b) Coverage and accuracy in benchmarking cePPI against reference EcoCyc PPI data set; p-value (ΣLLS vs other scores) computed by Mann-Whitney test and corrected for multiple hypotheses testing using Benjamin and Hochberg. (c) Fraction of interaction partners from E. coli CEPs vs non-CEPs; p-value by Kolmogorov-Smirnov test. (d) Average gene ontology (GO) semantic similarity of the putatively interacting CEPs from this study (CE network) vs other published studies. P-value significance (computed by Student's t-test) between CE network vs other studies for biological processes or cellular component GO annotations are shown with asterisk. (e) Box plots showing the Pearson correlation coefficients of the genetic interaction (GI) profiles for pairs of cePPIs compiled from the indicated E. coli studies vs randomly (Rand) drawn protein pairs; p-value by Student's t-test. (f) Physically interacting CEPs hypersensitive to drugs targeting the CE or different processes; p-value by Student's t-test. (g) Assessment of performance measures by comparing the area-under-the curve (AUC) for interactions that are deemed high confidence (HC) and/or medium confidence (MC). “All” refers to HC+MC cePPIs, as well as from previously reported large-scale (Butland et al., 2005; Hu et al., 2009, and Rajagopala et al., 2014) E. coli studies (using 5-fold cross validation), in benchmarking against cePPIs from curated EcoCyc protein complexes.