Artificial Intelligence Meets Toxicology
Igor V. Tetko, Günter Klambauer, Djork-Arné Clevert, Imran Hussain Shah, Emilio Benfenati · Chemical Research in Toxicology · 2022
Modern machine learning (ML), which is the basis of artificial intelligence (AI), has heavily impacted all fields of science including chemistry. 1 New ML methods based on deep neural networks and representation learning tend to provide a higher predictive quality in comparison to traditional computational methods.The interpretation of such models using explainable AI approaches allows us to identify atomic and fragment contributions responsible for the activity of molecules, and such methods are actively developing now. 2 New methods for extraction of information based on image 3 and text processing 4 are on the rise and will allow, in the future, to produce large sets of more accurately extracted and comprehensively annotated data from literature.Modern ML and AI methods are becoming increasingly popular in computational toxicology and are seen as a promising solution to meet the principles of the 3Rs concept, which calls for the replacement, reduction, and refinement of animal testing.5 Despite decades of research, descriptor-based quantitative structure-activity relationship (QSAR) methods based on traditional ML methods are still widely used in this field.However, there is a gap between cutting edge theoretical developments, featured in dedicated deep learning conferences or specialized ML journals and the end users who could practically use them.Moreover, the new methods are data hungry and require many examples to develop high-quality models.Emerging solutions, including the use of meta-and transfer-learning approaches that allow to pretrain models on a large corpus of chemical data of different Views expressed in this editorial are those of the authors and not necessarily the views of the ACS.