DDU predictor

Author: Marek Gierlinski ([email protected])

This repository provides scripts to prepare input data and run drug propertied predictor for DDU compounds, using a large training data (provided).

Usage

The scripts in this repository require an R package drupr, which contains all relevant code.

One of the provided training sets can be used (in training_data directory).

• lombardo.csv - to find volume of distribution, Vd
• master.csv - to find solubility, permeability, etc.

Each of the training sets is accompanied by an info file with information about columns that are used as id, response and rejected.

Merge and convert Excel files

There should be three Excel files as input: with pH7.4 descriptors (main file), with pH2 descriptors and Moka designations. In the first step these files are merged into one CSV file using mergeph. For example:

bin/mergeph.R -d excel_files/647/647_structures2_export_descriptors.xlsx -p excel_files/647/647_structures2_export_pH2.xlsx -o test_data/647_set.csv

creates a file test_data/647_set.csv which can be used for predictions.

Run predictor

Predictor is called as in this example:

rfpred.R -t training_data/master.csv.gz -i training_data/master.info.csv -s test_data/647_set.csv -o results/647 -m 500

The -m parameter is important and needs to be specify. It tells the predictor to ignore variables (columns) with less than m non-missing values.