GPD

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.

Install

conda create -n GPD
source activate GPD
module load cuda/11.3.0
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch
conda install -c conda-forge mdtraj
conda install -c anaconda networkx

Example

cd example/
sh submit_example_2_fixed.sh  (simple example)
sh submit_example_1.sh (fix some residue positions)

Output example:

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.

Training the GPD model

Dataset

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

Training the GPD model

train/train_encoder3.py Its training lasted 1 days and utilized 1 NVIDIA 40G A100 GPUs