Current deep learning methods, such as AlphaFold and Chai, can often create antibody-antigen (AbAg) structures with high confidence and accuracy. However, these methods often fail to predict the correct antibody binding site, placing the antibody incorrectly on the antigen and converging on repeatedly predicting the same redundant binding mode. Varying seeds and diffusion samples does not guarantee that a diverse set of binding modes explored. BepiPocket is a simple approach for integrating B-cell epitope prediction tools, BepiPred-3.0 and DiscoTope-3.0, to guide antibody-epitope restraints during Chai-1 structure prediction. This vastly increase diversity of explored antibody binding sites as well as allowing Chai-1 to find substantialy more accurate AbAg structures with high confidence.
BepiPocket is a tool developed by the Health Tech section at Technical University of Denmark (DTU). The code and data can be used freely by academic groups for non-commercial purposes. If you plan to use these tools for any for-profit application, you are required to obtain a license (contact Morten Nielsen, morni@dtu.dk).
We recommend Python version 3.11 for this tool.
$ conda create -n bepipocket python=3.11
$ conda activate bepipocket
$ conda install pip
$ conda install bioconda::anarci
Additionally, if you want to create MSAs as input for Chai-1, using the scripts in './scripts', you'll need an MMseqs2 installation.
$ conda install bioconda::mmseqs2
First, download requirements.txt file. Then,
$ pip install -r requirements.txt # install package dependencies
$ pip install git+https://github.com/mnielLab/BepiPocket-1.0.git
$ pip install --no-deps git+https://github.com/Magnushhoie/DiscoTope-3.0.git
After installation, BepiPocket can be run with the script 'main.py'.
Using the example fasta file, this code snippet does 4 runs, each producing 5 structures. The first run is without any antibody-epitope restraint, just using seed 0. The second, third and and fourth runs are made using antibody-epitope restraints constructed with BepiPred-3.0.
python main.py -i ./examples/2j88_ag_A_ab_L_H.fasta -o ./example_output/bepipocket -pred bepipocket -nr_runs 4
Here is another example using, DiscoPocket with MSA input
python main.py -i ./examples/2j88_ag_A_ab_L_H.fasta -o ./example_output/discopocket -pred discopocket -msa_directory ./examples/msa_pipeline/3chai1_pqt -nr_runs 4
If you found this tool useful in your research, please cite:
Pocket Restraints Guided by B-Cell Epitope Prediction improves Chai-1 Antibody-Antigen Structure Modeling