IgFold: Fast, accurate antibody structure prediction from deep learning on massive set of natural antibodies.

“IgFold is a structure prediction tool for heavy-light chain antibody structures. IgFold utilizes a novel approach to train a natural language model on 558M natural antibody sequences to predict structures that are more accurate than AlphaFold (and faster!)

Running IgFold on COSMIC²

To run IgFold:

  • Create a plain text file that has two entries: H & L.
  • Upload this file using the web browser upload
  • Click through tool creation. (There are no options to specify)
  • Then click submit

The job will finish in 20-30 minutes.

Output files

IgFold will output a predicted antibody structure (“my_antibody.pdb”) that will appear on the output data page.

Example IgFold FASTA file contents:

>H
EVQLVQSGPEVKKPGTSVKVSCKASGFTFMSSAVQWVRQARGQRLEWIGWIVIGSGNTNYAQKFQERVTITRDMSTSTAYMELSSLRSEDTAVYYCAAPYCSSISCNDGFDIWGQGTMVTVS

>L
DVVMTQTPFSLPVSLGDQASISCRSSQSLVHSNGNTYLHWYLQKPGQSPKLLIYKVSNRFSGVPDRFSGSGSGTDFTLKISRVEAEDLGVYFCSQSTHVPYTFGGGTKLEIK

Reference

Fast, accurate antibody structure prediction from deep learning on massive set of natural antibodies
Jeffrey A. RuffoloLee-Shin ChuSai Pooja MahajanJeffrey J. Gray