Vabs-Net

Vabs-Net: Pre-Training Protein Bi-level Representation Through Span Mask Strategy On 3D Protein Chains

The repository is an official implementation of Pre-Training Protein Bi-level Representation Through Span Mask Strategy On 3D Protein Chains

Environment installation

cd Vabs-Net
bash env.sh

Small molecule binding site dataset

In this paper, we also proposed a new small molecule binding site dataset, which can be found in BindingSiteDataset4VabsNet

version

We found a slight difference for EC and GO evaluation and updated the re-evaluated results in the arxiv version. The difference occurred because there was an issue with the calculation of the f1max.

Example of reproduction

Pre-train

cd Vabs-Net
bash reproduce_pretrain.sh

Finetune

cd Vabs-Net
bash reproduce_finetune.sh

Cite

If you want to cite this paper:

@inproceedings{zhuangpre,
  title={Pre-Training Protein Bi-level Representation Through Span Mask Strategy On 3D Protein Chains},
  author={Zhuang, Wanru and Song, Jia and Li, Yaqi and Lu, Shuqi and others},
  booktitle={Forty-first International Conference on Machine Learning}
}