One of the ways to use AI for drug discovery is to have models that predict the structure of molecules and how they might interact in chemical pathways. Scientists at the EPFL research institute in Switzerland have made an AI system which predicts nuclear magnetic resonance signatures of molecules. This technology can be used to help design new drugs. The research was published in Nature Communications earlier this week. Nature Communicationsvolume 9, Article number: 4501 (2018)
The machine learning system was trained using a dataset of calculated chemical shifts for many thousands of compounds. The trained machine learning system is available as open source software.
To predict the NMR signature of a crystal with nearly 1,600 atoms, our technique – ShiftML – requires about six minutes; the same feat would have taken 16 years with conventional techniques