Repertoire Builder: high-throughput structural modeling of B and T cell receptors

Repertoire Builder (https://sysimm.org/rep_builder/) is a method for generating atomic-resolution, three-dimensional models of B cell receptors (BCRs) or T cell receptors (TCRs) from their amino acid sequences. It is currently capable of handling batches of up to 10(4) sequences in approximately 30 minutes. This performance was achieved by applying a multiple sequence alignment extension technique originally developed for phylogenetic analysis to the template selection problem of complementarity determining region (CDR) loops. Under comparable conditions, average all-atom root-mean square deviations (RMSDs) from experimentally-determined structures of CDRH3 loops in BCRs were significantly lower than tested third-party high-throughput modeling methods, including ABodyBuilder, PigsPro, and LYRA. For TCRs, similar trends were observed when Repertoire Builder was compared with TCRmodel and LYRA. We also found that Repertoire Builder model errors were, in general, lower than those produced by our earlier Kotai Antibody Builder, even when CDRH3 loop refinement was used. However, in a subset of cases, which could be distinguished by poor Repertoire Builder scores, refinement by Kotai Antibody Builder or Rosetta Antibody, both of which utilize extensive structural sampling, improved the third heavy chain CDR (CDRH3) RMSD on average. Taken together, these results indicate that the MSA extension approach used by Repertoire Builder resulted in a favorable balance between speed and accuracy when compared to alternative methods. Furthermore, we conclude that more sensitive scoring, rather than extended structural sampling, is needed to further improve the accuracy of BCR and TCR modeling.