Figure 6. The GBM model has higher accuracy and lower error count than blindBLAST.

(A) Comparison of blindBLAST (red) and GBM (blue) accuracy in assigning CDR sequences to clusters. Error counts and accuracies are averaged over each of 3 repeats (in the 3-repeat 10-fold CV scheme) for both GBM and blindBLAST (see Methods). (B) Comparison of the number of erroneously assigned clusters in blindBLAST and GBM error count. The GBM model universally lowers error count.