Predicting protein interface residues from residue conservation between interacting and non-interacting groups

It would be great if interface residues, which are responsible for protein-protein interactions, could be determined from the amino acid sequence. Sequence Harmony, Multi-RELIEF, SDPpred and GroupSim+ConsWin are used to predict interface residues from residue conservation between interacting and non-interacting groups. Therefore, a dataset based on Fungal Orthologous Groups is used. Different analysis methods are used for Sequence Harmony. Some interesting differences are found in these methods. Sequence Harmony, Multi-RELIEF, SDPpred and GroupSim+ConsWin perform significantly better then a random prediction of interface residues. The statistical analyses are performed with care: Quantile-Quantile plots are created for each True Positive Rate to calculate accurate confidence intervals. The interface predictions are significant better then random but somewhat disappointing. Also, large deviations in predicting individual FOG interfaces are found for every method and high correlations between predicting individual FOG interfaces are only found between Sequence Harmony and MultiRELIEF (Figure 14).