Machine learning for natural extracts

Abstract

Cheminformatics Natural products and their derivatives continue to be an important source of drug candidates because of their structural diversity and wide-ranging biological activities, which are unmatched by synthetic compounds. Natural products are generally complex mixtures with chemical constituents that are not well characterized. Reher et al. report a nuclear magnetic resonance–based machine-learning tool, SMART, for rapid structural analysis of major constituents from crude natural extracts and for the discovery of new natural products. For example, SMART automatically characterized a cyanobacterial extract mixture and isolated a new chimeric macrolide, symplocolide A; it also dereplicated several known natural products. The proposed cheminformatic tool paves the way for new computer-aided approaches to natural product drug discovery. J. Am. Chem. Soc. 142 , 4114 (2020).