​
Antibody sequence design

AntiFold is an antibody-specific model developed by OPIG which generates diverse sequences which fold into a structure. It’s fine-tuned from ESM-IF1 and allows for structure-based antibody design and optimization, with the antigen in context. It excels in sequence recovery across CDRs and demonstrates superior performance to previous methods in predicting antibody-antigen binding affinity, including ProteinMPNN and AbMPNN.

AntiFold on Tamarind has been modified to support sampling sequences from nanobody inputs and specifying custom CDR indices (IMGT numbered CDRs are used by default).

​
Inputs

  • PDB File
  • Heavy Chain ID
  • Light Chain ID
  • Whether you’d like to include the antigen in context
  • Regions to mutate (ex. CDR1, CDRH1, FW1)

​
Outputs

  • Fasta file with generated sequences
  • CSV with log probabilities for each residue at each position (less negative means more likely mutation)

Try AntiFold