About antiberty

[YiningWang2]

This is an interesting work. I saw you used pretrained model, AntiBERTy to represent the protein sequence and this package had been installed by pip. But I didn't see where it is used in the code. Could you give me any idea?

[jeffreyruffolo]

I ended up including AntiBERTy within the pre-trained weights of IgFold to ensure proper AntiBERTy are inputted into IgFold, so it doesn't actually show up in the code explicitly. You can see the model being called here: https://github.com/Graylab/IgFold/blob/c31ee2ece3d567c2efea07185850a263b628372b/igfold/model/IgFold.py#L141

In case you want to use AntiBERTy as a standalone, it is available on PyPI as you mentioned. The documentation needs more detail, but after loading the model it should behave like a standard HuggingFace BERT model. https://pypi.org/project/antiberty/

[YiningWang2]

Okay，I see. Thanks so much. Very helpful answer!

[BartoszJanuszNA]

I ended up including AntiBERTy within the pre-trained weights of IgFold to ensure proper AntiBERTy are inputted into IgFold, so it doesn't actually show up in the code explicitly. You can see the model being called here:

https://github.com/Graylab/IgFold/blob/c31ee2ece3d567c2efea07185850a263b628372b/igfold/model/IgFold.py#L141

In case you want to use AntiBERTy as a standalone, it is available on PyPI as you mentioned. The documentation needs more detail, but after loading the model it should behave like a standard HuggingFace BERT model. https://pypi.org/project/antiberty/

Hello @jeffreyruffolo, I am trying to fine tune AntiBERTy using the standalone package you mentioned here, but it lacks tokenizer. Do I assume correctly that I can create my own using resources/vocab.txt from IgFold repo? The encoding ids didn't changed between IgFold and standalone AntiBERTy, did they?

[jeffreyruffolo]

Hello, yes that is correct. AntiBERTy uses the BertTokenizer from HuggingFace with that vocab file.

[BartoszJanuszNA]

Thank you for quick response!