IgFold is an antibody structure prediction method that uses embeddings from the AntiBERTy (1) and extensively uses invariant point attention (2).
Details
- Nodes initialized from hidden layer embeddings
- Edges initialized from all interresidue attention matrices from each layer
- Interchain embeddings across H and L chain set to zero
- Uses the two invariant point attention sets, the second of which consists of three layers with unique weights
- Includes some distilled training data from AlphaFold2
- Error estimation also uses Invariant point attention
- A study found it to generate unrealistic bond lengths and angles, particularly in CDRH3
- Results are competitive with AlphaFold Multimer
1.
Ruffolo JA, Gray JJ, Sulam J. Deciphering antibody affinity maturation with language models and weakly supervised learning. 2021; Available from: https://arxiv.org/abs/2112.07782
2.
Ruffolo JA, Chu L-S, Mahajan SP, Gray JJ. Fast, accurate antibody structure prediction from deep learning on massive set of natural antibodies. Nature Communications. 2023;14(1). Available from: https://doi.org/10.1038/s41467-023-38063-x