Figure 1.
The overlap between the largest positive patch and the real RNA-binding interface in three different RBPs.
(A) L1 ribosomal protein (1mzp), (B) rotavirus nonstructural protein (1knz), and (C) tymovirus coat and capsid binding protein (1ddl). The blue region represents the largest positive patch, yellow is the real binding interface (calculated as described in the Materials and Methods section), and green denotes the overlap between the extracted patch and the real interface. Notably, in (A) there is a large overlap (0.9) between the positive patch and the interface, while a very small overlap (0.05) was observed in (C).
Figure 2.
Illustration of the three positive electrostatic patches in the aspartyl tRNA synthetase (1asy).
The largest patch is colored blue, the second largest patch is magenta, the third largest patch is cyan, and the negative patch is colored red. Interestingly, for the tRNA-binding proteins the protein binds via both the positive and the negative electrostatic patches.
Figure 3.
Patch size and surface potential of RBPs, DBPs, and NNBPs.
Patch size is plotted against the average surface potential for all RBPs (black diamonds) compared to DNA-binding proteins (crosses) and non-NA binding proteins (open diamonds). As can be noticed, a large number of NNBP are characterized by relative large patch size.
Figure 4.
ROC plots illustrating the SVM results for RBPs classification.
In black, RNA-binding proteins versus non-NA-binding proteins (AUC = 0.90); in red, RNA-binding proteins versus non-NA-binding proteins with large patches (AUC = 0.88); in green, RNA-binding proteins versus non-NA-binding proteins when including only the electrostatic patch properties (AUC = 0.81).
Table 1.
Summary of SVM results for different classifiers.
Table 2.
Summary of the discriminating features selected by SVM-RFE.
Figure 5.
The largest electrostatic patch mapped on the protein structure of two RRM domains.
(A) The U2 snRNP A′ from the U2B″–U2A′ complex (1a9nA) and (B) the Y14 protein from the Y14-Magoh complex (1rk8A). Blue represents the largest electrostatic patch and green the RRM motif as defined by pfam. For the RNA-binding RRM domain the largest electrostatic patch overlaps the RNA-binding interface, while no overlap is observed between the largest electrostatic patch and the protein-protein interface of the Y14 protein. Notably, the largest positive patch is much smaller in the latter case.
Figure 6.
Spearman correlation coefficient values (ρ) calculated for each one of the 40 features comparing tRNA vs. all RBPs.
The features are colored by group ( detailed numbers are given in Dataset S1): Dark blue represent features related to the largest positive patch, in red are features related to the whole protein, in green are cleft-patch related features, and in cyan are the “other patches” features. The protein feature and the features related to the secondary electrostatic patches showed the highest CC with a positive sign, denoting that these features are greater in the tRNA group.
Figure 7.
Multiclass SVM analysis for 3 subgroups.
(A) mRNA, (B) rRNA, and (C) tRNA. Each protein in each of the subgroups was tested against the three different classifiers. For each subgroup, the SVM results for the mRNA classifier are shown in the most left column, results for the rRNA classifier in the middle column, and for the tRNA classifier in the right column. SVM results are color-coded: red representing high positive results and shaded blue representing low negative results (see color bar).
Table 3.
A table summarizing the multi-SVM results for 3 subclasses of RNA-binding proteins: tRNA, rRNA, and mRNA.
Figure 8.
The correlation between the patch-interface overlap and the discriminate value obtained from the SVM classifier.
As illustrated, the prediction power of the algorithm depends on the success in identification of the functional interface.