This repository provides the code for the paper "Latent Retrieval Augmented Generation of Cross-Domain Protein Binders". RADiAnce is a diffusion model that generates protein binder by leveraging retrieval-augmented generation techniques.
You can create and activate the environment by:
conda env create -f env_cuda117.yaml
conda activate RADiAnceFor evalutation, you need to install PyRosetta, plaease refer to PyRosetta Installation for details.
Suppose all the datasets are downloaded below /path/to/data/.
All data are saved under /path/to/data/peptide. We set environment variable export PREFIX=/path/to/data/peptide.
Download:
# LNR
wget https://zenodo.org/records/13373108/files/LNR.tar.gz?download=1 -O ${PREFIX}/LNR.tar.gz
tar zxvf ${PREFIX}/LNR.tar.gz -C $PREFIX
# PepBench
wget https://zenodo.org/records/13373108/files/train_valid.tar.gz?download=1 -O ${PREFIX}/pepbench.tar.gz
tar zxvf $PREFIX/pepbench.tar.gz -C $PREFIX
mv ${PREFIX}/train_valid ${PREFIX}/pepbench
# ProtFrag
wget https://zenodo.org/records/13373108/files/ProtFrag.tar.gz?download=1 -O ${PREFIX}/ProtFrag.tar.gz
tar zxvf $PREFIX/ProtFrag.tar.gz -C $PREFIXProcessing:
python -m scripts.data_process.peptide.pepbench --index ${PREFIX}/LNR/test.txt --out_dir ${PREFIX}/LNR/processed --remove_het
python -m scripts.data_process.peptide.pepbench --index ${PREFIX}/pepbench/all.txt --out_dir ${PREFIX}/pepbench/processed
python -m scripts.data_process.peptide.transform_index --train_index ${PREFIX}/pepbench/train.txt --valid_index ${PREFIX}/pepbench/valid.txt --all_index_for_non_standard ${PREFIX}/pepbench/all.txt --processed_dir ${PREFIX}/pepbench/processed/
python -m scripts.data_process.peptide.pepbench --index ${PREFIX}/ProtFrag/all.txt --out_dir ${PREFIX}/ProtFrag/processedAll data are saved under /path/to/data/antibody. We set environment variable export PREFIX=/path/to/data/antibody.
- SAbDab (Downloaded at Sep 24th)
mkdir ${PREFIX}/SAbDab
wget https://opig.stats.ox.ac.uk/webapps/newsabdab/sabdab/archive/all/ -O ${PREFIX}/all_structures.zip
wget https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabdab/summary/all/ -O ${PREFIX}/SAbDab/summary.csvProcessing:
# process
python -m scripts.data_process.antibody.sabdab --index ${PREFIX}/SAbDab/summary.csv --out_dir ${PREFIX}/SAbDab/processed
# split
python -m scripts.data_process.antibody.split --index ${PREFIX}/SAbDab/index.txt --rabd_summary ${PREFIX}/RAbD/rabd_summary.jsonl
python -m cal_topk --config ./configs/contrastive/calculate/topk_protein.yaml --ckpt path/to/contrastive_vae --gpu 0 --save_dir your_pathImportant: You need to change the rag database path in the train configs(e.g../configs/LDM-rag/train_pt.yaml) to your own path.
Training of the full RADiAnce requires 8 GPUs with 80G memmory each.
GPU=0,1,2,3,4,5,6,7 bash ./scripts/train.sh ./configs/contrastive/train_pt.yaml # train contrastive autoencoder
GPU=0,1,2,3,4,5,6,7 bash ./scripts/train.sh ./configs/LDM-rag/train_pt.yaml # train rag diffusion model, you should also modify the path of the rag database and checkpoints in the config file
# peptide
python generate.py --config configs/test/test_pep.yaml --ckpt /path/to/checkpoint.ckpt --gpu 0 --save_dir ./results/pep
# antibody
python generate.py --config configs/test_rag/test_abxx.yaml --ckpt /path/to/checkpoint.ckpt --gpu 0 --save_dir ./results/abxx