Fig. 4

Sequence similarity differences vs. structural similarity differences. Each point represents a CDR3β, and its coordinate represents the difference in sequence similarity, CDRdist, (x-axis) and structural similarity, RMSD, (y-axis) between the most similar sequence in its STCRDab structural class and the most similar sequence in any other STCRDab CDR3β structural class. A negative number indicates that the CDR is more similar to the neighbor in its structural class than to the other neighbor; i.e., left indicates more similarity in sequence and bottom more similarity in structure to an in-class CDR than to an out-of-class one. Color represents the distance between the CDR and the in-class neighbor. Structural superpositions highlight example relationships between a target CDR (blue), its in-class neighbor (green), and its other neighbor (orange). a Sequence-based classification is correct, but actually does not yield the best structural match. b-d Classification is correct, and yields (b) similar, (c) better, and (d) much better structural matches. e-f Classification according to STCRDab classes is incorrect but is actually consistent with the relative structural similarity: the other structure is (e) more similar or (f) about the same, though in (e) neither structure is particularly similar. g Classification is incorrect, and the closest sequence in the same structural class is more similar to that in the other class