DrugPred_RNA by Illimar Rekand, Ruth Brenk University of Bergen, 2020

To run DrugPred_RNA, you need Python 3.6 and R version 4.0.3 and UCSF DOCK 3.6 (http://dock.compbio.ucsf.edu/)

The following Python packages are necessary to run this software:

• rdkit
• biopython
• openbabel

The following packages are also necessary in R (but should be autoinstalled the first time you run it):

• ggplot2
• SHAPforxgboost
• xgboost
• fastshap

Download the files

Download all files from the DrugPred_RNA repository on GitHub and maintain the folder structure

Preparing the conda environment to run DrugPred_RNA

It's possible to create a conda environment with the necessary packages from the .yml-file. To do this, run

conda env create --file DrugPred_RNA_env.yml

You should now be able to run

conda activate drugpred_RNA

Unzip the database containg drug molecules for docking and prepare environment

1. create the environment variable DOCK_BASE_3_6 and point it to the folder containing DOCK 3.6 (e. g. setenv DOCK_BASE_3_6 /path_to_dock_folder/dockenv)

2. create the environment variable DOCK_BASE and point it to the same folder as $DOCK_BASE_3_6

3. create the environment variable DrugPred and point it to the foler containing the DrugPred scripts

4. at the following folders to your path: $DOCK_BASE/bin/, $DOCK_BASE/private, $DOCK_BASE/etc, $DOCK_BASE/scripts, $DOCK_BASE/bin/Linux, $path.

(e. g. set path = ($DOCK_BASE/bin/ $DOCK_BASE/private $DOCK_BASE/etc $DOCK_BASE/scripts $DOCK_BASE/bin/Linux $path))

Downloading and preparing files for DrugPred_RNA

STEP 1.1: cd into the main directoy of DrugPred_RNA. Here, you should have the folders "scripts" and "input_csvs".

STEP 1.2: Create a .txt file, and name it "PDB_ids.txt". Add the four-letter PDB ids you want DrugPred_RNA to predict in this document, separating entries with a comma. ie. "2yie, 5c45" etc.

STEP 1.3: Run: python scripts/retrieve_pdb_structures.py -i PDB_ids.txt

Step 1.4:

cd into the directory named "PDB-FILES", which contains all the downloaded .cif-files with the matching PDB-id's.

Step 1.5:

Run: python ../scripts/clean_up_pdb_files.py -f .cif

Step 1.6:

In the main directory there is now a file name "entries.csv", which contains the ids of all the entries which DrugPred_RNA will evaluate. The file consists of the PDB id and the three letter ligand code, separated by an underscore. Note that if a binding site contains a metal ion, separate entries will be created for each individual metal, ie. "2yie_FMN", "2yie_FMN_MG".

Docking, calculating and predicting the binding sites with DrugPred_RNA

STEP 2.1:

cd into the directory PDB-FILES

STEP 2.2 Run: python ../scripts/drugpred.py -s -1.3 --density 2

This may take a while as DrugPred will dock each binding site

Analyzing the output from DrugPred_RNA

When the script is finished, cd back to the main directory. You should now have the following files:

• "Predictions.csv", which contains the predicted outcomes of the assessed binding sites. Evaluated binding sites are evaluated as druggable (>= 0.5) or less druggable (<0.5)
• "descriptor_values.csv", which contains the calculated descriptor valus for each assessed binding site.
• In the "cleaned"-directory, subdirectories for each assessed binding site is made. Here, cd into the docking subdirectory to find:
  1. The binding site which was used for docking, with the suffix "_cofac.pdb"
  2. The original ligand of the binding site, with the suffix "_lig.pdb"
  3. The superligand file with the suffix "exp_atoms_modified.pdb". This is the superligand created by DrugPred_RNA. Atoms with "TRP"-labelled residues are solvent exposed, "ASP" are buried.
  4. The atoms which are in direct contact with the superligand has the suffix "_1.0_bs.pdb"
• A directory named "ind_shap_plots". In here, SHAP plots for each assessed binding site is stored. These show the top 5 strongest influences descriptors have on the predicted outcome of each binding site.