Graph based deep learning method to design protein sequences.There are twon main models which are TMPNN-alpha and TMPNN-beta.
- TMPNN-alpha is only used for inverse folding (also called fixed backbone design).
- TMPNN-beta v1.0 is trained to not only capture the residue identity information but the residue topology information. We add a conditional random field after the encoder block.
- TMPNN-beta v2.0 add an IPA module from OpenFold to explicit model the backbone information.
Author : Bo Zhang
E-mail : zhangbo777@sjtu.edu.cn