Improving energetic feature selection to classify protein-protein interactions

Protein-protein interactions (PPIs) are known for its important role in diverse biological processes. One of the crucial issues to understand and classify PPI is to characterize their interfaces in order to discriminate between transient and permanent complexes. The stability of protein-protein interactions depends on the energetic features of interaction surfaces. This work explores the surfaces of complex interaction classified as permanent and transient, in order to find those energetic features that can differentiate between both type of complexes. We claim that the number of energetic features and their contribution to the interactions can be key factors to predict between transient and permanent interactions. Moreover, the features used can be adjusted according to the size of the complex studied. We evaluate different classifiers to predict these interactions, using a set of 298 complexes extracted from databases of protein complexes -in terms of their known three-dimensional structure-, and which were already classified as transient or permanent. As a result, we obtained an improved accuracy up to 86.6% when using SVM with kernel linear.