DeltaVina

Tutorial

This is the latest version of tutorial for Δ_vinaXGB, old version can be found here
First Part and Second Part of this tutorial is about installation of dependencies and set up the Δ_vinaXGB.
Third Part and Forth Part are the dataset and examples of applying Δ_vinaXGB scoring function to rescore the protein-ligand binding affinity.

---

1. Setup

• Download deltaVinaXGB
• Creat environment
  

  make Makefile create_environment
conda activate DXGB
make Makefile requirements
  

• Install all packages

  conda install -c conda-forge xgboost=0.80.0
conda install -c rdkit rdkit=2019.03.1
conda install -c openbabel openbabel
conda install -c conda-forge setuptools
  

• Install source code

  python setup.py install
  

---

2. All Dependencies

• MGLTools and MSMS
• A modified version of AutoDock Vina
• R with randomForest library
• Set the environment variable

• MGLTools and MSMS

MGLTools are used for preparing PDBQT file for Vina. MSMS is for calculating solvent accessible surface area. Both of them can be downloaded from MGLTools website.

http://mgltools.scripps.edu/downloads

Note

• MGLTools can be installed by

  tar -xvzf mgltools_x86_64Linux2_1.5.6.tar.gz
cd mgltools_x86_64Linux2_1.5.6/
./install.sh

  Remember use correct mgltools name (Linux or Mac) when install mgltools.
• After downloaded msms_i86_64Linux2_2.6.1.tar.gz
  

  mkdir msms
tar -xvzf msms_i86_64Linux2_2.6.1.tar.gz -C msms
cd msms
  

  Copy msms.x86_64Linux2.2.6.1 to be msms
  

  cp msms.x86_64Linux2.2.6.1 msms
  

• In msms folder, there is a script pdb_to_xyzr. Change the line
  numfile = "./atmtypenumbers" to be numfile = "YourPATHofddeltaVinaXGB/DXGB/atmtypenumbers" atmtypenumbers file we used can be found in deltaVinaXGB/DXGB directory
• Test pdb_to_xyzr

  pdb_to_xyzr 1crn.pdb > 1crn.xyzr

  If it doesn't work, try

  ./pdb_to_xyzr 1crn.pdb > 1crn.xyzr

  If error: nawk: command not found, change nawk to awk in pdb_to_xyzr (line 31)

• A modified version of AutoDock Vina

A fork of Autodock Vina has been modified to output the features in score_only mode.

https://github.com/chengwang88/vina4dv

After download, extract files from zip file.If directory name is vina4dv_master, change it into vina4dv.

• R with randomForest library (Needed for Δ_vinaRF₂₀)

Most recent version of R is recommended.

https://cran.r-project.org/

Note

• Install randomForest in R by
  

  install.packages('randomForest')

• Set the environment variable

If you have the dependencies installed already. Several environment variables need to be set in .bashrc (Linux) or .bash_profile (macOS) file in your home directory. An example is given below. You can modify the path based on your case. In this example, all softwares are installed under /home/jl7003 directory.

Example

# path for MSMS
export PATH=$PATH:/home/jl7003/msms/

# set mgltool variable (if mac, should change mgltools_x86_64Linux2_1.5.6 into your downloaded mac version)
export PATH=$PATH:/home/jl7003/mgltools_x86_64Linux2_1.5.6/bin/
export MGL=/home/jl7003/mgltools_x86_64Linux2_1.5.6/
export MGLPY=$MGL/bin/python
export MGLUTIL=$MGL/MGLToolsPckgs/AutoDockTools/Utilities24/

# set vina dir
export VINADIR=/home/jl7003/vina4dv/build/linux/release/
br>

After set all environment variables, open a new window to enable all setups.

---

3. Dats Set

Before calculating features, three structure inputfiles are needed:

pdbid_ligand.mol2/sdf --> ligand structure file
pdbid_protein.pdb --> protein structure
pdbid_protein_all.pdb --> protein with water molecules structure file
Examples for structure files can be found in Test_2al5 directory.
Note: these three files are needed to run predictions. All files should include hydrogens. If protein files with water molecules are not available, just copy the original protein.pdb to protein_all.pdb.

If features have been already calculated, only features files are needed:

Input.csv--> Input feature file
Example for Input.csv can be found in Test directory.

---

4. Run Model

After all of above have been set up, the example can be run in deltaVinaXGB/DXGB.

conda acivate DXGB
cd DXGB

You can check the help by

python run_DXGB.py --help

The script can be run for one complex by

python run_DXGB.py --runfeatures --datadir ../Test_2al5 --pdbid 2al5 --average

--runfeatures is feature calculation, default is to calculate all features, --datadir is for structure files datadir, --pdbid is for structure pdbid, can be other type of index. --average is to calculate average scores from 10 models.

Or it can also be run by providing a list of protein-ligand complex with input features as in Input.csv

python run_DXGB.py --datadir ../Test --average


Default is to predict scores for provided structures. If you want to get scores with explicit water molecules, and optimized ligands:

python run_DXGB.py --runfeatures --datadir ../Test_2al5 --pdbid 2al5 --water rbw --opt rbwo --average

--water is for consideration of water effect, rbw is to consider both receptor-bound water and bridging water molecules, --opt is for optimization, rbwo is to optimize ligand in no water, bridging water, and receptor-bound water environemnts.

The calculated features will be saved in Input.csv file, and the predicted scores will be saved in score.csv file. If you want to get deltaVinaRF scores as well, add --runrf.