LightRoseTTA - Pytorch

Installation

1. Clone the package

git clone https://github.com/psp3dcg/LightRoseTTA.git
cd LightRoseTTA

2. Create conda environment using LightRoseTTA-env.yml file

# create conda environment for LightRoseTTA
conda env create -f LightRoseTTA-env.yml

3. Download and Install the biological packages

$ bash install_bio_package.sh
$ copy bio_package(blast, csblast, psipred) to your_path/LightRoseTTA/msa_feat

4. Download the Uniref30[46G], BFD[272G] and pdb100[over 100G] datasets

$ python download_datasets.py

Test datasets

Download the test datasets from google drive

Testing steps

# run the testing python file
python test_script.py [FASTA_folder_path] [data_write_path] [Uniref30_dataset_path]  [pdb100_dataset_path] [BFD_dataset_path] [model_file_path]
	
-FASTA_folder_path: the path of folder containing FASTA files
-data_write_path: the path of folder to write generated data
-Uniref30_dataset_path: the path of Uniref30 dataset
-pdb100_dataset_path: the path of pdb100 dataset
-BFD_dataset_path: the path of BFD dataset
-model_file_path: the path of model file

For example,
# for general proteins
python test_script.py ./Orphan25_fasta ./Orphan25_data ./Uniref30_2020_06 ./pdb100_2021Mar03 ./BFD ./weights/LightRoseTTA.pth

# for antibodies
python test_script.py ./Antibody_fasta ./Antibody_data ./Uniref30_2020_06 ./pdb100_2021Mar03 ./BFD ./weights/LightRoseTTA-Ab.pth

The output "*.pdb" files are located in "data_write_path/predict_pdb" (e.g. Orphan25_data/predict_pdb)

Training steps

# prepare the training data provided by LightRoseTTA_preprocess_train_data
(a) download the LightRoseTTA_preprocess_train_data.zip and unzip it
(b) prepare the ".fasta" files and corresponding ".pdb" files
(c) cd "LightRoseTTA_preprocess_train_data" folder and generate data following the README.md

# run the training python file
(a) download the LightRoseTTA_train_data.zip and unzip it
(b) cd "LightRoseTTA_preprocess_train_data" folder
(c) python LightRoseTTA_train.py -dataset [training_data_path]
ps: training_data_path should include "raw" folder and "processed" folder

References

X Wang, et al., LightRoseTTA: High-efficient and Accurate Protein Structure Prediction Using an Ultra-Lightweight Deep Graph Model, bioRxiv 10.1101/2023.11.20.566676 (2023).