Various Benchmarking Experiments Implemented in QSPRpred

This repository contains the code used to conduct the case study experiments in the paper titled "QSPRpred: a Flexible Open-Source Quantitative Structure-Property Relationship Modelling Tool" and is intended as a reference implementation of a streamlined QSPR modelling worfklow for the purpose of comparison of models and other aspects of QSPR modelling.

Install Dependencies

Different dependencies are required for different experiments. The following should be sufficient for all experiments:

Note: In order to ensure compatibility of all software in both experiments, use Python 3.10.

pip install -r requirements.txt # for software dependencies

Run Experiments

Settings for each experiment is stored as a module under settings. You just point to the module you wish to run with the QSPBENCH_SETTINGS environment variable and run the main.py script. For example, to run the multi-task regression experiments from the first experiment you can issue the following command:

QSPBENCH_SETTINGS='settings.experiment_1.multitask_reg' python main.py

Some settings files define other variables that can be used to customize behaviour. Check the settings you intend to run for more details.

Analyze Results

For each experiment, a Jupyter notebook that reads and analyzes the output data is available in the root of the repository (i.e. analyze_experiment_1.ipynb).