AB-Gen A tool for antibody library design
Manual
This is the source code for AB-Gen, a tool for antibody library design. It can efficiently explore the antibody space and design CDRH3 sequences that fulfill multi-property constraints.
Installation
Clone the repository
git clone git@github.com:charlesxu90/ab-gen.git
cd ab-gen
Create environment
conda env create -f environment.yml
conda activate ab-env
Download dataset and pretrained models
git lfs pull
Install netMHCIIpan
- Download NetMHCIIpan 4.1 as shown in the guide.
- Modify
NMHOME
in/netMHCIIpan-4.1/netMHCIIpan
file to be your unzipped folder, e.g.
setenv NMHOME ~/Desktop/netMHCIIpan-4.1
Test if netMHCIIpan-4.1 works properly:
python -m agent.scoring.MHCAffinity
Running the code
Generate from pretrained models
# Prior model
python generate.py --model_path data/models/Prior_10_1.704.pt --out_file result/prior/10k_samples.txt
# Agent HER2 model
python generate.py --model_path data/models/Agent_her2_final.pt --out_file result/agent/agent_oas_her2/10k_samples.txt
# Agent MPO model
python generate.py --model_path data/models/Agent_mpo_final.pt --out_file result/agent/agent_oas_mpo/10k_samples.txt
Reproduce the figures
Run jupyter-lab
in command line. Then open each jupyter notebook to reproduce the figures.
License
This code is licensed under MIT License.