Figure 4. CDRs that are misclassified have a relatively large dihedral angle distance to their cluster exemplars.

Density of dihedral angle distances between the query loops and their cluster exemplar loops. For most loop lengths and types (e.g., H1-13, H2-9, L1-15, L3-8, L3-9, and L3-10), the misclassified distribution has more density at larger dihedral angle distances with respect to the correctly classified distribution. The skewedness indicates that for many loops, if a query CDR is distant to its corresponding structural exemplar, then it is more likely to be incorrectly classified using the blindBLAST method.