Abstract

Machine learning methods such as quantitative structure-activity/property relationship (QSAR/QSPR) models that are based on regression or classification approaches can accelerate the design of advanced materials by a few orders of magnitude. Their potential has been demonstrated in a variety of different disciplines ranging from chemical and biological sciences, drug discovery to engineering challenges over the course of the last decades. In the presented use cases, we will employ physicochemical properties or theoretical molecular descriptors of chemicals as predictors. First, the QSAR/QSPR models try to identify a relationship between the molecular structures and properties of interest (e.g., the effect on the corrosion behaviour of lightweight alloys like aluminium and magnesium) in a given dataset of chemicals. In a second step, the trained models can be employed to predict the properties of untested chemicals. After a short introduction to the theoretical basics, deeper insights into the creation of QSAR/QSPR models, their application and associated pitfalls will be presented.