Graphormer-based Protein Design (GPD) model deploys the Transformer on a graph-based representation of 3D protein structures and supplements it with Gaussian noise and a sequence random mask applied to node features, thereby enhancing sequence recovery and diversity. The performance of GPD model was significantly better than that of state-of-the-art model for ProteinMPNN on multiple independent tests, especially for sequence diversity.
One can use pip to directly install our package
pip install fair-GPD
For 40 series gpus, we recommend use the following methods to install. For the current pip installed pytorch may have some errors with 40 series gpus.
conda create -n GPD
source activate GPD
conda install pytorch==1.12.1 -c pytorch
conda install -c conda-forge mdtraj==1.9.9
conda install -c anaconda networkx==3.1
Note that GPD could be used with cuda, you can install the cudatoolkit package according to your own gpu version.
Also, one could use our given environment.yml file to create an environment
conda env create -f environment.yml
One can use our given requirements.txt file for pip installation
pip install -r requirements.txt
cd test/
sh submit_example_2_fixed.sh (simple example)
sh submit_example_1.sh (fix some residue positions)
outputs/example_1_outputs/1tca.fasta
> predicted model_0 acc: 0.3501577287066246 length: 317
APTGAAPPLTLPPATLRAQLAAKGASPEDLKNPVLILHGPGTDGAEDFAGFLVRLLKSKGYTPAYVDPDPN
ALDDIADDLEALALAAKYLAAGLGNKPFNVITHSLGGVALLTALAYHPELRDKIKRVVLVSPLPTGSDSLR
ALLAANTLRLLQFLSVKGSALDDAARKAGALTPLVPTTVIGHANDPLHYPTSLGSPASGAYVPDARVIDLY
SVYGPDFTVDHAEAVFSSLVRKALKAALTSSSGYARASDVGKSLRVSDPAKDLSAEQREAFLNLLAPAAAA
IANGKTGNACPPLPPEYLPAAPGAKGAGGVLTP
> predicted model_1 acc: 0.334384858044164 length: 317
APTGEPLPLLLPDATLLANVEADGADIDEVTNPVLLLHGLGSDGEEALGASLVALLKALGYTPLGVDPDPN
YTDDILDDAQALAAAARALAAGLGNKPLLVVGHSLGGVVVLLALRYNPALADLIASVILVAPAPRGSSEAR
PLIAAKILRPEDFLLLYGSALADALRAAGLDVPLVPTTVIDSADDPLHSPNALLSAESAAYVPGGTVVDLS
DIFGPDFTVSHAGAVLSPFLRKLLEAALASPTGVPREEDVGASLLDLDLAADLTAEERAAALNALAAYAAR
IAAGARFNAYPALPPELVPAAKGATDAAGTLKP
- acc is recovery. Recovery was the proportion of the same amino acids at equivalent position between the native sequence and the designed sequence
- length is the length of designed sequence.
the training dataset and three test datasets: The GPD model was trained using the CATH 40% sequential non-redundancy dataset, with a split ratio of 29868:1000:103 for the training, validation, and testing sets, respectively. We further evaluated the performance of GPD using 39 de novo proteins, including 14 de novo proteins that exhibit significant structural differences from proteins belonging to natural folds.
- data/cath-dataset-nonredundant-S40-v4_3_0.pdb is CATH 40% sequential non-redundancy dataset downloaded from http://download.cathdb.info/cath/releases/all-releases/v4_3_0/non-redundant-data-sets/cath-dataset-nonredundant-S40-v4_3_0.pdb.tgz
- data/sc103 is 103 single chain proteins
- data/denovo39 is 39 de novo proteins
- data/denovo14 is 14 de novo proteins
The pre-trained weights and scripts for calculating graph features
The script is for training the model using graph features. The output of this script includes model weights and the detailed training process including loss and recovery for each epoch. Its training lasted 1 days and utilized 1 NVIDIA 40G A100 GPUs
The one-line bash command that allows you to employ GPD for designing your proteins.