search

THUNLP-MT / dyMEAN

This repo contains the codes for our paper "End-to-End Full-Atom Antibody Design"
https://arxiv.org/abs/2302.00203
MIT License
89 stars 8 forks source link

Is it possible to use a predicted antibody structure to initialize X_i^(0)? #5

Open semal opened 1 year ago

semal commented 1 year ago

Excellent work, and very thx for open source. This is my question. If Using tools like Igfold or alphafold to predict antibody, then use this structure to initialize X. Of course for the [MASK] residue using a random residue to replace when using the predict tools. Have you try something like this?

kxz18 commented 1 year ago

Hi, I didn't try that. To do that, maybe we need to train a new model. However, if you're using Igfold to predict the structure, I think directly letting dyMEAN to do the structure prediction might achieve similar or even better performance, as compared in the paper.