Figure 2. A flowchart of supervised and semi-supervised learning methods used to predict the effect of nsSNPs on PPIs.

A. Shown is the protocol of training both supervised and semi-supervised methods for the 3-class problem (mutations of detrimental/neutral/beneficial effects). The semi-supervised learning method depicted here is the random-forest self-learning classifier. B. Feature representation of each nsSNP was calculated by taking energy differences between the wild-type and mutant complexes. The mutant PPI complex was modeled by FoldX using as a template the structure of wild-type complex. C. During the prediction stage, the classifier assigns a new nsSNP to one of the classes.