Understanding complex biological or chemical systems requires the ability to observe them in real time in a non-invasive and non-destructive manner.
Raman spectroscopy provides a method for such observations by utilizing the interaction of light and matter through Raman scattering. This spectroscopy method is powerful as it can provide molecular ‘fingerprints’ of both organic and inorganic substances, including live cells. Especially, this possibility highlights the potential for using Raman spectroscopy in a variety of applications, such as monitoring the biological process of fermentation, the study of chemical reactions or visualizing drug delivery processes in real time.
As the Raman signal usually is weak and sparse, there is a need for high precision instruments. This has limited the method’s use outside of laboratory settings. Recently, however, advances have been made to the instruments recording the spectra such that they are now capable of obtaining Raman spectra at a high spatio-temporal resolution (ie. high resolution Raman ‘videos’). These advancements render the classical signal processing pipelines infeasible due to error accumulation across each step of the pipeline and the relatively low speed with which the analysis can be performed.
This project aims to advance the statistical methods required to analyze high resolution Raman spectra by utilizing an end-to-end joint statistical modeling approach. This is done by combining a semi-parametric model for peak detection in Raman maps for analyzing the spectra themselves with an amortized inference scheme aiming to learn a fast-to-compute forward mapping directly from the observed spectra to quantities of interest in the Raman spectra.
The successful outcome of this project will be a fully automated procedure for analyzing Raman maps with high spatio-temporal resolution in reasonable computing time by formulating the entire Raman spectroscopy analysis pipeline as a single statistical model optimized with modern machine learning methods. This will contribute to the advancement of Raman spectroscopy in realistic settings and in research within analysis of spectral data.
Image: by Associate Professor Mikkel N. Schmidt