Innopolis Researchers Create AI for Decoding Molecular Mass Spectra
Researchers at Innopolis University have developed a computational AI framework for analyzing tandem mass spectra — a kind of molecular "fingerprint."
The neural network was trained on more than 1 million spectrum–molecule pairs and maps both data types into a single vector space. The solution automatically analyzes data and produces a list of the most suitable molecules within seconds: processing a single spectrum takes around 0.09 seconds when searching a database of hundreds of thousands of structures, and in 90.5% of cases the correct structure appears in the top hundred results.
The development will help pharmacists, chemists, and laboratory staff identify the structure of unknown compounds faster. A patent has been obtained for the technology.
