Comparison of different methods for the prediction of critical residues in protein interfaces using a BID derived dataset as described in Tuncbag et al. [18].

a<p>Predictions were performed using PCRPi <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0012352#pone.0012352-Assi1" target="_blank">[7]</a> with an expert BN trained in a Ab+ dataset that does not include the crystal structure of the c2 fragment of streptococcal protein G in complex with the Fc domain of human Ig (PDB code 1fcc).</p>b<p>Values were obtained running FoldX <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0012352#pone.0012352-Guerois1" target="_blank">[19]</a> with default parameters and a ddG<sub>binding</sub> cut-off of 2.0 Kcal.mol<sup>−1</sup> (i.e. residues were considered critical if upon mutation to Ala, predicted ddG<sub>binding</sub>≥2.0 Kcal.mol<sup>−1</sup>).</p>c<p>Precision, recall, and F1 score values taken from Tuncbag et al. <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0012352#pone.0012352-Tuncbag1" target="_blank">[18]</a>.</p>