Learning protein structural similarity from sequence alone. Our preprint can be found here

DeepBLAST is a neural-network based alignment algorithm that can estimate structural alignments. And it can generate structural alignments that are nearly identical to state-of-the-art structural alignment algorithms.

DeepBLAST can be installed from pip via

pip install deepblast

To install from the development branch run

pip install git+https://github.com/flatironinstitute/deepblast.git

The pretrained DeepBLAST model can be downloaded here.

The TM-align structural alignments used to pretrain DeepBLAST can be found below

• Training data
• Validation data
• Testing data

See the Malisam and Malidup websites to download their datasets.

See the wiki on how to use DeepBLAST and TM-vec for remote homology search and alignment. If you have questions on how to use DeepBLAST and TM-vec, feel free to raise questions in the discussions section. If you identify any potential bugs, feel free to raise them in the issuetracker

If you find our work useful, please cite us at

@article{morton2020protein,
  title={Protein Structural Alignments From Sequence},
  author={Morton, Jamie and Strauss, Charlie and Blackwell, Robert and Berenberg, Daniel and Gligorijevic, Vladimir and Bonneau, Richard},
  journal={bioRxiv},
  year={2020},
  publisher={Cold Spring Harbor Laboratory}
}

@article{hamamsy2022tm,
  title={TM-Vec: template modeling vectors for fast homology detection and alignment},
  author={Hamamsy, Tymor and Morton, James T and Berenberg, Daniel and Carriero, Nicholas and Gligorijevic, Vladimir and Blackwell, Robert and Strauss, Charlie EM and Leman, Julia Koehler and Cho, Kyunghyun and Bonneau, Richard},
  journal={bioRxiv},
  pages={2022--07},
  year={2022},
  publisher={Cold Spring Harbor Laboratory}
}