DeepTernary

https://github.com/youqingxiaozhua/deepternary
Admin
DeepTernary is an SE(3)-equivariant graph neural network designed to predict the 3D structures of...
Tokens:3,072
Snippets:31
Trust Score:6.5
License:Apache-2.0
Update:2 months ago
Show doc for...
Tokens:
Raw
### Install DockQ

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/DockQ/README.md

Clone the repository and compile the tool using make.

```bash
git clone https://github.com/bjornwallner/DockQ/
cd DockQ
make
```

--------------------------------

### Install PyTorch dependencies

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/README.md

Command to install a specific version of PyTorch with CUDA support.

```bash
pip install torch==2.3.1+cu121 --index-url https://download.pytorch.org/whl/cu121
```

--------------------------------

### Start Distributed Training

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/README.md

Initiates training on 2 GPUs using the specified configuration file.

```Bash
bash tools/dist_train.sh deepternary/configs/protac.py 2
```

--------------------------------

### View Help Page

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/DockQ/README.md

Display command-line arguments and usage instructions.

```bash
./DockQ.py -h
```

--------------------------------

### Run DockQ

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/DockQ/README.md

Execute the scoring function on a model and native PDB file.

```bash
./DockQ.py examples/model.pdb examples/native.pdb
```

--------------------------------

### Display TernaryDB directory structure

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/README.md

Visual representation of the expected dataset folder hierarchy.

```text
TernaryDB
├── pdbs
│   ├── 1A2Y_A_C_PO4
│   │   └── gt_complex.pdb
│   ...
├── MGD
│   ├── test.txt
│   ├── train_clusters.json
│   ├── train.txt
│   └── val.txt
└── PROTAC
    ├── test.csv
    ├── train_clusters.json
    ├── train.txt
    └── val.csv
```

--------------------------------

### Data Directory Structure

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/README.md

Expected directory layout for the TernaryDB dataset.

```text
data/TernaryDB
├── preprocessed  # PREPROCESSED_PATH
│   └──  pdb2311_merge  # extracted from preprocessed/pdb2311_merge.tar 
├── ligand_ideal  # IDEAL_PATH, extracted from ligand_ideal.tar
└── pdbs  # PDB_PATH, extracted from pdb2311_merge.tar.xz
```

--------------------------------

### Evaluate pre-trained models

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/README.md

Commands to download, extract, and run evaluation on the PROTAC test set.

```bash
# download the pre-trained checkpoints and unbound structures
wget https://github.com/youqingxiaozhua/DeepTernary/releases/download/v1.0.0/output.zip

# unzip data
unzip output.zip

# evaluate on the PROTAC test set
python predict_cpu.py output/checkpoints/PROTAC

# MGD need to process the test data, see `TernaryDB/MGD/test.txt`
# python predict_cpu.py output/checkpoints/MGD
```

--------------------------------

### Run permutation analysis

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/DockQ/README.md

Use the -perm flags to automatically test all possible chain permutations to find the best scoring alignment.

```bash
bash$ ./DockQ.py examples/1A2K_r_l_b.model.pdb examples/1A2K_r_l_b.pdb -native_chain1 A B -perm1
1/2 AB -> C 0.00972962403319
Current best 0.00972962403319
2/2 BA -> C 0.476267208558
Current best 0.476267208558
Best score ( 0.476267208558 ) found for model -> native, chain1:BA -> AB chain2:C -> C
***********************************************************
*                       DockQ                             *
*   Scoring function for protein-protein docking models   *
*   Statistics on CAPRI data:                             *
*    0.00 <= DockQ <  0.23 - Incorrect                    *
*    0.23 <= DockQ <  0.49 - Acceptable quality           *
*    0.49 <= DockQ <  0.80 - Medium quality               *
*            DockQ >= 0.80 - High quality                 *
*   Reference: Sankar Basu and Bjorn Wallner, DockQ:...   *
*   For comments, please email: bjornw@ifm.liu.se         *
***********************************************************
Model  : examples/1A2K_r_l_b.model.pdb
Native : examples/1A2K_r_l_b.pdb
Best score ( 0.476267208558 ) found for model -> native, chain1:BA -> AB chain2:C -> C
Number of equivalent residues in chain A 248 (receptor)
Number of equivalent residues in chain B 196 (ligand)
Fnat 0.491 26 correct of 53 native contacts
Fnonnat 0.103 3 non-native of 29 model contacts
iRMS 1.988
LRMS 7.300
CAPRI Medium
DockQ_CAPRI Acceptable
DockQ 0.476
```

```bash
./DockQ.py examples/1A2K_r_l_b.model.pdb examples/1A2K_r_l_b.pdb -native_chain1 A B -perm1 -perm2
```

```bash
./DockQ.py examples/1EXB_r_l_b.model.pdb examples/1EXB_r_l_b.pdb -native_chain1 A B C D -perm1
```

```bash
./DockQ.py examples/1EXB_r_l_b.model.pdb examples/1EXB_r_l_b.pdb -native_chain1 A B C D -perm1 -perm2
```

--------------------------------

### Evaluate multi-chain docking models

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/DockQ/README.md

Commands to specify chain groupings for multi-chain complexes. The second group is automatically formed from remaining chains if not explicitly specified.

```bash
./DockQ.py examples/1A2K_r_l_b.model.pdb examples/1A2K_r_l_b.pdb -native_chain1 A B -model_chain1 A B -native_chain2 C -model_chain2 C
```

```bash
./DockQ.py examples/1A2K_r_l_b.model.pdb examples/1A2K_r_l_b.pdb -native_chain1 A B
```

```bash
./DockQ.py examples/1A2K_r_l_b.model.pdb examples/1A2K_r_l_b.pdb -native_chain1 C
```

```bash
./DockQ.py examples/1A2K_r_l_b.model.pdb examples/1A2K_r_l_b.pdb -native_chain1 A B -model_chain1 A B
```

```bash
./DockQ.py examples/1A2K_r_l_b.model.pdb examples/1A2K_r_l_b.pdb -native_chain1 A B -model_chain1 B A
```

--------------------------------

### Run PROTAC Prediction

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/README.md

Executes a PROTAC prediction task using unbound settings and multiple seeds.

```Bash
python predict.py \
    --task PROTAC \
    --name 5T35-no-correct \
    --lig output/protac22/5T35_H_E_759/ligand.pdb \
    --p1 output/protac22/5T35_H_E_759/unbound_protein1.pdb \
    --p2 output/protac22/5T35_H_E_759/unbound_protein2.pdb \
    --unbound-lig1 output/protac22/5T35_H_E_759/unbound_lig1.pdb \
    --unbound-lig2 output/protac22/5T35_H_E_759/unbound_lig2.pdb \
    --lig1-mask output/protac22/5T35_H_E_759/unbound_lig1.pdb \
    --lig2-mask output/protac22/5T35_H_E_759/unbound_lig2.pdb \
    --outdir ./results/protac_case \
    --seeds 40
```

--------------------------------

### Run MGD Prediction

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/README.md

Executes an MGD prediction task using bound settings and a single seed.

```Bash
python predict.py \
    --task MGD \
    --name 6HR2-case \
    --lig data/TernaryDB/pdbs/6HR2_F_E_FWZ/ligand.pdb \
    --p1 data/TernaryDB/pdbs/6HR2_F_E_FWZ/protein1.pdb \
    --p2 data/TernaryDB/pdbs/6HR2_F_E_FWZ/protein2.pdb \
    --outdir ./results/mgd_case
```

--------------------------------

### BibTeX Citation for DeepTernary

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/README.md

Use this BibTeX entry to cite the DeepTernary paper in academic publications.

```BibTeX
@article{xueDeepTernary2025,
  title = {{{SE}}(3)-Equivariant Ternary Complex Prediction towards Target Protein Degradation},
  shorttitle = {{{DeepTernary}}},
  author = {Xue, Fanglei and Zhang, Meihan and Li, Shuqi and Gao, Xinyu and Wohlschlegel, James A. and Huang, Wenbing and Yang, Yi and Deng, Weixian},
  year = {2025},
  month = jul,
  journal = {Nature Communications},
  volume = {16},
  number = {1},
  pages = {5514},
  publisher = {Nature Publishing Group},
  issn = {2041-1723},
  doi = {10.1038/s41467-025-61272-5},
  urldate = {2025-07-22},
  langid = {english}
}
```

--------------------------------

### Fix Residue Numbering

Source: https://github.com/youqingxiaozhua/deepternary/blob/master/DockQ/README.md

Use the Perl script to align residue numbering between model and native files.

```bash
scripts/fix_numbering.pl model.pdb native.pdb
```
Tokens:
Raw