Cambridge, MA-based startup Kebotix has created machine learning software that learns material chemistry from three-dimensional models of molecules with known properties in order to design novel compounds.
Kebotix feeds the molecular models to a neural network that learns a statistical representation of their properties, which can devise new examples aligned with existing models; a second network screens out undesirable designs, then a robotic system tests the chemical structures of the remaining models.
The outcomes are input back into the machine learning channel so it can yield results closer to target properties.
The Massachusetts Institute of Technology's Klavs Jensen said the use of such automation in chemistry "won't replace the expert, but you'll be able to do things a lot faster."
From Technology Review
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA