< Back to Article

Quantitative Protein Localization Signatures Reveal an Association between Spatial and Functional Divergences of Proteins

Figure 2

Performance of PLAST in identifying known co-localized proteins.

(A) Probability distributions of the P-profile dissimilarity scores (dp) between interactors and between non-interactors detected by affinity-purification mass spectrometry (AP-MS) or yeast two-hybrid screening (Y2H). (M = medians, μ = means of the distributions; P-values from two-sided permutation tests for differences in means or medians.) (B) An example of PLAST search result obtained from using 19S proteasomal base subunits as query proteins. The mean dp between the query proteins and all other proteins are shown as red (known subunits) or gray (other ORFs) vertical lines. Most of the red lines have low dp values, indicating that they are placed at the top of the search result. (Black line graph = precisions, red line graph = recalls, black dashed line = decision threshold at optimum F1-score, black box = magnified region, parenthesis = number of known subunits.) (C) Performances of subunit searches obtained from using different numbers of query proteins randomly selected from known subunits of a proteasome (left) or cytosolic ribosome (right). For each query protein number, we tested max(100, number of all possible combinations) random combinations of query proteins, and computed the mean value of these tested combinations (parentheses = numbers of known subunits). (D) Normalized F1-score differences between P-profileSVM and UCSF annotation for a catalog of 197 protein complexes. Some of the complexes with the highest F1-score differences are highlighted. (Red line = the mean of the normalized differences, which is also the test statistic used in the paired t-test between the F1-scores of these two methods; gray areas = statistically insignificant differences with P>0.001; red text = unadjusted P-value obtained for the paired t-test.) (E) Bonferroni-adjusted P-values obtained from one-sided, paired t-tests between the F1-scores of of all the possible pairs of profiling/annotation methods (F1A vs. F1B).

Figure 2