Antigen-Specific Antibody Design and Optimization with Diffusion-Based Generative Models for Protein Structures (NeurIPS 2022) [Paper]

conda env create -f env.yaml -n diffab
conda activate diffab

The default cudatoolkit version is 11.3. You may change it in env.yaml.

Protein structures in the SAbDab dataset can be downloaded here. Extract all_structures.zip into the data folder.

The data folder contains a snapshot of the dataset index (sabdab_summary_all.tsv). You may replace the index with the latest version here.

Trained model weights are available here.

HDOCK is required to design CDRs for antigens without bound antibody frameworks. Please download HDOCK here and put the hdock and createpl programs into the bin folder.

PyRosetta is required to relax the generated structures and compute binding energy. Please follow the instruction here to install.

Ray is required to relax and evaluate the generated antibodies. Please install Ray using the following command:

pip install -U ray

5 design modes are available. Each mode corresponds to a config file in the configs/test folder:

Below is the usage of design_pdb.py. It samples CDRs for antibody-antigen complexes. The full list of options can be found in diffab/tools/runner/design_for_pdb.py.

python design_pdb.py \
	<path-to-pdb> \
	--heavy <heavy-chain-id> \
	--light <light-chain-id> \
	--config <path-to-config-file>

The --heavy and --light options can be omitted as the script can automatically identify them with AbNumber and ANARCI.

The below example designs the six CDRs separately for the 7DK2_AB_C antibody-antigen complex.

python design_pdb.py ./data/examples/7DK2_AB_C.pdb \
	--config ./config/test/codesign_single.yml

HDOCK is required to design antibodies for antigens without bound antibody structures (see above for instructions on installing HDOCK). Below is the usage of design_dock.py.

python design_dock.py \
	--antigen <path-to-antigen-pdb> \
	--antibody <path-to-antibody-template-pdb> \
	--config <path-to-config-file>

The --antibody option is optional and the default antibody template is 3QHF_Fv.pdb. The full list of options can be found in the script.

Below is an example that designs antibodies for SARS-CoV-2 Omicron RBD.

python design_dock.py \
	--antigen ./data/examples/Omicron_RBD.pdb \
	--config ./config/test/codesign_multicdrs.yml

python train.py ./configs/train/<config-file-name>

🚧 Coming soon...