BCL_Mol2D_benchmark

This repo contains scripts to perform the benchmark and sensitivity analysis in the paper BCL::Mol2D – a robust atom environment descriptor for QSAR modeling and lead optimization

1. Download and install BCL

The Bio Chemical Library (BCL) is a software package that provides a suite of cheminformatics tools that allow construction of quantitative structure-activity-relation (QSAR) models for virtual screening, pharmacophore mapping, and drug design. BCL is free of charge for non-comercial use.

Follow the instruction on this page to obtain access to the source code of BCL.

2. Benchmark different descriptor configurations in QSAR tasks

In the benchmark folder:

Make dataset bin files

• Download the smi string file of nine PubChem datasets here. Convert the smi string files to SDF files with 3D conformation using OpenBabel and Corina as described in the method section of the paper. Store the sdf files in the data folder.

• In the data folder, make the bin data files by running:

./make_bin_files.sh

Train the QSAR models

Train and benchmark the QSAR models with different descriptor configurations by running this script:

qsar_train.sh

The computational jobs can be performed locally or on computer cluster via srun (see the script)

3.Sensitivity analysis

In the sensitivity_analysis folder:

Compute the decrement and increment dervatives of different atom environments for each active and inactive compound by running:

./compute_sensitivity_scores.sh