Determining molecular properties with differential mobility spectrometry and machine learning

• DOI: 10.1038/s41467-018-07616-w
• Publication Date: 2018-11-26T11:53:01Z
• Authors: Stephen W. C. Walker; Ahdia Anwar; Jarrod M. Psutka; Jeff Crouse; Chang Liu; J. C. Yves Le Blanc; Justin Montgomery; Gilles H. Goetz; John S. Janiszewski; J. Larry Campbell; W. Scott Hopkins
• Keywords: 0303 health sciences; 03 medical and health sciences; Science; Q; Article; 3. Good health

The fast and accurate determination of molecular properties is highly desirable for many facets chemical research, particularly in drug discovery where pre-clinical assays play an important role paring down large sets candidates. Here, we present the use supervised machine learning to treat differential mobility spectrometry - mass data ten topological classes We demonstrate that gas-phase clustering behavior probed our experiments can be used predict candidates' condensed phase properties, such as cell permeability, solubility, polar surface area, water/octanol distribution coefficient. All these measurements are performed minutes require mere nanograms each examined. Moreover, by tuning gas temperature within spectrometer, one fine tune extent ion-solvent separate subtly different geometries discriminate molecules very similar physicochemical properties.