Physiological and evolutionary gene features are predictive of genetic interaction degree. (a) Gene features are significantly correlated with negative genetic interaction degree. We measured the Pearson correlation coefficients between gene feature values and negative genetic interaction degree for 3,456 non-essential S. cerevisiae genes. Error bars show 95% confidence intervals. A complete set of features and their correlations is given in Table 1; see Materials and methods for descriptions of gene features. (b) Overview of the regression tree model for genetic interaction degree. An ensemble of 100 decision trees was built from bootstrap samples of genes. Combinations of values of features are represented as paths from the root to the leaves of a tree. Internal nodes each split data (sets of genes) according to values for a single feature; leaf nodes are associated with predicted genetic interaction degrees. (c) Scatter plot of negative genetic interaction degree and degrees predicted by the bagged decision tree model on held-out genes shows the significant relationship between predicted and actual degrees (Pearson's r = 0.80, P < 10-324). FD, fitness defect; PPI, protein-protein interaction; SM, single mutant.