Community-based Deep Learning and Antibody H3 Loop Modeling
Create a conda environment with this command. It will automatically create an
environment named h3_loop. You could also optionally provide the name of
the environment.
conda env create -f environment.yml [-n <name>]If conda takes too long, it's recommended to use mamba instead.
mamba env create -f environment.yml [-n <name>]Then run the following command. (Thanks to Dr.Marc Bianciotto)
conda activate commat
python3 setup.py install
pip3 install git+https://github.com/NVIDIA/dllogger#egg=dllogger
conda install bioconda::anarci
apt-get install openmpi-binThe run_inference.sh script contains an example execution command.
The FASTA file should follow the format {pdbname}_{Hchain}_{Lchain}_{antigen_chain}. For example, if the PDB ID is 7sn1 and only the H chain is present in the FASTA file, the input file should be named 7sn1_H_#_#.fa. If both the H and L chains are present, the file should be named 7sn1_H_L_#.fa. If the antigen chain is A, the file name should be 7sn1_H_L_A.fa.
When you provide a FASTA file and a seed size, the script generates different structures up to the specified seed size and outputs them in the output_folder/relaxed directory with ranking labels. For instance, if the file is named relaxed/7sn1_H_L_#_1.pdb, it indicates that this structure has a ranking of 1. If you want to quickly obtain unrelaxed structures and use a separate relaxation tool, you can use the outputs in the output_folder/unrelaxed directory.
Due to technical issues, the AF2rank tool could not be included, but you can download and use it from ColabDesign if needed.
This related SO question describes the differences between the two commands.
All code, except for the code in the "src/galaxylocalopt" directory, is licensed under the MIT license. Code in the "src/galaxylocalopt" directory is licensed under the CC BY-NC-ND 4.0 license.