CavityPoser: Mining For Bound-Like Conformations of RNA Using a Binding Cavity Screening Approach.

• Python and Python packages: numpy, pandas and scikit-learn v0.21.2

• openbabel

• rdock

• PyMOL. For first time PyMOL users: you will need a PyMOL license file, as PyMOL is a commercial software.

git clone git@github.com:karoka/CavityPoser.git
cd cavity_mining/
git clone --depth=1 git@github.com:atfrank/RNAPosers.git
cd RNAPosers/
make clean
make
cd ..
cp RNAPosers/bin/featurize bin/
rm -rf RNAPosers

Main script is src/cavityPoser.sh. It requires a pdb file containing RNA 3D structure as input. Example:

cd test/
./../src/cavityPoser.sh receptor.pdb

The predicted cavities and their corresponding scores are written in file receptor/predicted_pockets.txt.

./../src/cavityPoser.sh

It takes the following arguments:

• path-to-pdb-file: input pdb file to perform cavity mining on.
• scalar: 1 or 0, corresponds to different version of models (1: scalar; 0: vector). Currently the scalar version is faster and has better performance. Default: 1
• rdock-param-file: rdock parameter file. 3 parameter files are provided under params/ folder. Default: blind_docking_de_novo.prm.
• working-dir: working directory. Default: a sub-directory in current working directory. The sub-directory is named as the input pdb-file name without extension. e.g. If the input filename is receptor.pdb, then a folder named receptor/ will be created if not existed in current working directory and all operations will be performed there.
• tag: an arbitrary identifier of the structure. Default: "decoy".
• output-file: output file path and name.

cd training/

• Required python environment: pandas, scikit-learn, xgboost

• Use pre-trained models: python train.py 1

• Re-train models: python train.py 0

• Get ranking table: python get_rank 1 and python get_rank 1

• Output: results/rank_testset1.csv and results/rank_testset1.csv

• RNAPosers(In revision): Chhabra, Sahil, Jingru Xie, and Aaron T. Frank. "RNAPosers: Machine Learning Classifiers For RNA-Ligand Poses." bioRxiv (2019): 702449.