Automating data analysis for vibrational spectroscopy - using AI for chemical identification
05 Apr 2023
- Rosie de Laune



An artificial intelligence model that identifies the structure of a molecule from its infrared spectrum has been developed by PhD students Guwon Jung and Son Gyo Jung with their supervisor Jacqui Cole.


Graphic showing the use of a convolutional neural network (CNN) to determine molecular structure from an infrared spectrum.


Infrared spectroscopy is a method commonly used by chemists to identify the functional groups present in a molecule, and therefore its overall structure. Different functional groups absorb infrared light of different frequencies, reducing the intensity of the beam passing through the sample and creating a unique fingerprint in the resulting spectrum.

Analysing these spectra is usually done manually, by comparing your result with spectra from the literature. This process can be challenging, as the peaks may shift depending on the functional group's surrounding environment or overlap with one another.

In a recent study, published in Chemical Science, students Guwon Jung and Son Gyo Jung, along with their supervisor Professor Jacqui Cole who holds a joint appointment at ISIS and the University of Cambridge, developed a model that uses neutral networks to identify 37 different functional groups from infra-red spectra.

Guwon and Son are both based at the Research Complex at Harwell, with Guwon funded by STFC's Scientific Computing Department and Son funded by both ISIS and the University of Cambridge. Guwon trained and validated the model using a dataset of over 30,000 unique compounds, developing a tool that can be used by researchers for quick and easy chemical identification. 

Building on this tool, the group hope to expand the work to other forms of spectroscopy, including the vibrational spectroscopy that is captured by the inelastic neutron scattering instruments at ISIS, such as Tosca. So, in the future, data analysis for ISIS users could become a whole lot easier! 

Further information

The full paper can be found at DOI: 10.103​9/D2SC05892H

Contact: Cole, Jacqui (Cambridge Uni.,RAL,ISIS)