PyMolSAR: a Python-based toolkit to predict the activity and property of small molecules

Thesis (Master's)--University of Washington, 2018Machine learning is a powerful approach for generating Quantitative structure-activity relationships (QSAR) models to predict the property and biological activity of small molecules. However, building such models in Python is cumbersome for cheminformatics researchers as they must use several Python packages and undertake a sequence of modeling steps. For instance, use Python packages for calculating molecular descriptors and generating models. Therefore, a Python toolkit that integrates these Python packages and modeling steps will immensely benefit cheminformatics researchers. This work presents a Python toolkit, called PyMolSAR for building predictive structure-activity relationships models for small molecules. The functionality of PyMolSAR includes calculating 759 1D/2D molecular descriptors, data preprocessing, feature selection, training and evaluating predictive models. It is open-source and freely available on GitHub at https://github.com/BeckResearchLab/PyMolSAR