Please use this workflow to run a sequence humanisation experiment by providing a paired, non-human starting sequence. AbBFN2 will then be used to introduce mutations to the framework regions of the starting antibody, possibly using several recycling iterations. During sequence humanisation, appropriate human V-gene families to target will also be chosen, but can be manually set by the user too.
Once an experiment is completed, output can be downloaded in multiple formats. The PDB structures that we provide are folded using ImmuneBuilder.
Briefly, the humanisation workflow here uses the conditional generation capabilities of AbBFN2 in a sample recycling approach. At each iteration, further mutations are introduced, using a more aggressive starting strategy that is likely to introduce a larger number of mutations. As the sequence becomes more human under the model, fewer mutations are introduced at subsequent steps. Please note that we have found that in most cases, humanisation is achieved within a single recycling iteration. If the model introduces a change to the CDR loops, which can happen in rare cases, these are removed. For a detailed description of the humanisation workflow, please refer to the AbBFN2 manuscript.
Please also note that while we provide the option to manually select V-gene families here, this workflow allows the model to select more appropriate V-gene families during inference. Therefore, the final V-gene families may differ from the initially selected ones. Please also note that due to the data that AbBFN2 is trained on, humanisation will be most reliable when performed on murine or rat sequences. Sequences from other species have not been tested.