Fig. 1

Role correlations. Heatmap of role-role correlations for a subset of roles clustered according to the dendrogram clustering method in R. Roles are arranged according to their positions in a dendrogram (not shown) computed according to their mutual correlations. In particular, roles that are clustered together in the dendrogram will appear close to one another in the diagram; borders with high contrast correspond to divisions between higher-order clusters. This algorithm maximizes contrast in the heatmap at such boundaries and results in light-colored blocks of strongly correlated roles. High correlations along the diagonal correspond to highly conserved small sets of roles, e.g. subunits of a single protein complex, and all roles are fully correlated with themselves (ρ=1). While it is apparent from visual inspection of the blocks in the heatmap that there is an underlying structure to these role-role correlations, the actual nature of this structure can be nonapparent and difficult to characterize precisely. EvalCon uses machine learning to learn these structures from role-role correlations, thereby eliminating the need for an a priori characterization