FASTv2

Setup

Requirements

• Python == 3.8
• conda (recommended)

Install dependencies

The dependencies can be installed using the following command (conda required):

bash install.sh

NOTE: Please ensure that you have the repository for tfbio available for building from source before proceeding with the installation.

Preparing Datasets

Creating/linking datasets

How to download the dataset

The dataset can be donwloaded from this ftp link

Make sure to download all of the data to a data folder with name data in the current directory.

If dataset is not available, but raw docking files are present in folder

  python RawHDF5.py --dir=${DATA_DIR} --dataset=${DATASET} --sub-dataset=${SUBDATASET} --output-dir=${OUTPUTDIR} --pocket-type=${POCKET}

You may also run the following command:

  sbatch scripts/sbatch_dataset_curate -d ${DATASET} -s ${SUBDATASET} -o ${OUTPUTDIR} -p ${POCKET_TYPE}

Note:

1. Make sure all the files are in DATA_DIR
2. This feature is still under construction.

Hierarchial layout of DATA_DIR is shown below:

dengue (dataset)
├── denv2 (subset)
│   └── 2fom (pocket)
│      └── scratch
│          ├── dockHDF5
│          │   ├── dock_proc1.hdf5
│          │   └── ...
│          └── receptor.hdf5
└── denv3
    └── 3u1i
      └── scratch
          ├── dockHDF5
          │   ├── dock_proc1.hdf5
          │   └── ...
          └── receptor.hdf5

Note: Not all datasets will have a subset. This is an optional argument. Also, the data format should be in .hdf5 file format

Training

There are sample scripts in scripts directory.
A training job can be submitted with the following command.

sbatch scripts/sbatch_train.sh -c ${CONFIG_PATH} -t ${TAG} -d ${SAVE_DIR} 

If you are running on a local or interactive node, you can train with the command below.

python train.py --config_path=${CONFIG_PATH} --tag=${TAG}

Sample configurations can be found in the configs directory.

NOTE: You do not need to specify the training directory.
If run as above, a unique identifiable directory will be created in /usr/local/$USER with $TAG.
Of course, you can specify the save directory by adding --save_dir=${DIR} flag.

Resuming the previous training

If you want to resume the existing training, you only need to specify the existing directory through the --save_dir=${DIR} flag.

python train.py --save_dir=${DIR}

Evaluation

You can evaluate your trained model using the following command.

python eval.py --save_dir=${DIR}

NOTE: If you want to use a different configuration from the one previously used for training, you can specify through the --config_path=${CONFIG_PATH} flag.

Contact

Please contact Aditya Ranganath (ranganath2@llnl.gov) if you have any request.