2024 MIT R&D Conference: Track 1 - Space - Automating the Identification of Chemical Mixture Components with Machine Learning

• Video details
   

  Automating the Identification of Chemical Mixture Components with Machine Learning Brett McGuire Class of 1943 Career Development Assistant Professor, MIT Department of Chemistry Identifying the precise chemical makeup of complex mixtures is of interest in fields ranging from atmospheric chemistry to pharmaceutical development and quality control to my own field of astrochemistry.  A variety of analytical tools such as spectroscopy, mass spectrometry, nuclear magnetic resonance, and chromatography provide chemical "fingerprinting," which can, in theory, be used to identify these chemical components, but the sheer density of spectral features of different molecules that are often present in such readings can make unambiguous assignment to individual species challenging. Yet, the components are commonly chemically related due to the shared chemical evolution of the mixture. Therefore, along with investigating the analytical signals, analysis of the structural and chemical relevance of a molecule is an important consideration when determining which species are present in a given mixture. My group works primarily in applications of rotational spectroscopy, and thus, in this talk, I will present a method that combines machine-learning molecular embedding models with a graph-based ranking system to determine the likelihood of a molecule being present in a pure rotational spectrum based on the other known species, chemical priors, and spectroscopic information.  I'll present details on the process as well as demonstrate its utility on both laboratory mixtures and astrochemical observations from space.  Our work demonstrates that the chemical inventory can be identified with extremely high accuracy in a much more efficient manner than manual analysis.

Locked Interactive transcript