Require python>=3.7

Clone the repository and install it

git clone https://github.com/zw2x/glinter.git
cd glinter
pip install -e .

Manually install the following softwares and models

• torch>=1.6
• torch_geometric
• MSMS
• reduce
• hh-suite
• esm_msa1_t12_100M_UR50S

The taxonomy database can be found at

The following commands are executed in the repository glinter.

Replace the environment variables in set_env.sh and run.sh in the scripts directory by your installation paths and then run

source scripts/set_env.sh

There are two example pdb files in the examples directory. To predict the inter-protein heterodimeric contacts between 1a59A.pdb and 1a59B.pdb, run

scripts/build_hetero.sh examples/1a59A.pdb examples/1a59B.pdb examples/

To predict the inter-protein homodimeric contacts between 1a59A.pdb and 1a59B.pdb, and use 1a59B as the representative monomer (only build MSA for 1a59B), run

scripts/build_homo.sh examples/1a59A.pdb examples/1a59B.pdb examples/ 1a59B

Make sure the names of the receptor and ligand are different.

For heterodimers, the raw outputs are examples/1a59A:1a59B/1a59A:1a59B.out.pkl and examples/1a59A:1a59B/1a59B:1a59A.out.pkl (both directions). For homodimers, the raw output is examples/1a59A:1a59B/1a59A:1a59B.out.pkl.

For both cases, the final score (the probability that a pair of residue positions is in contact) is examples/1a59A:1a59B/score_mat.pkl. The ranked list of residue position pairs is examples/1a59A:1a59B/ranked_pairs.txt; the first column is the positions in the receptor, the second column is the positions in the ligand, the third column is the probability.

See alphafold/example_run.sh for using GLINTER given the predicted structures from AlphaFold.

Please use the uniclust database A3M_SpecBloc requires the header of each hit starts with tr| and contains OS=$TAX, $TAX is the taxonomy name.

Please use the model provided in the repo ckpts/glinter1.pt instead of the one from zenodo