Figure 3
From: Different protein-protein interface patterns predicted by different machine learning methods
Effects of three different algorithms in four methods. (A,B,C,D) each presents the results of SVM, random forest, logistic regression with lasso penalty, logistic regression with hierarchy interaction. The abscissa means numbers of residue pairs chosen to be interacting residue pairs in a dimer and the ordinate means numbers of correct predicted dimers as long as there is one truly interacting residue pair chosen correctly. In the legend, “original” means algorithm without EasyEnsemble and feature engineering, “EasyEnsemble” means algorithm that only using EasyEnsemble without feature engineering, “feature engineering and EasyEnsemble” represents the results obtained by both EasyEnsemble and feature engineering. In addition, “mean”, “median” and “weighted mean” indicated three ensemble methods that were used in both “EasyEnsemble” and “feature engineering and EasyEnsemble”.
