Fusion of spectroscopic and spectrometric data with machine learning approaches

• Applicant: Professor Dr. Stephan Seifert
• Subject Area: Analytical Chemistry
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 561375600

There are several analytical techniques for generating fingerprints that are used to classify and characterize biological samples, e.g. regarding their provenance or biological identity. These techniques are often based on spectroscopy, like nuclear magnetic resonance (NMR) and Fourier-transform near infrared (FT-NIR) spectroscopy or mass spectrometry, e.g. liquid chromatography mass spectrometry (LC-MS) or inductively coupled plasma mass spectrometry (ICP-MS). However, the data obtained by each of these methods only represent a small fraction of the complex composition of the samples and, hence, also the chemometric models for classification that are built from the respective data sets only rest on a very limited part of the existing differences. To build more robust and powerful classification models, we will develop and compare various data fusion approaches that combine complementary information from the different techniques. A key component of this will be the comprehensive characterization of class differences through the application of multiblock methods and novel machine learning approaches. The latter will be used for the selection of important variables and for the evaluation of their mutual impact on the outcome within and across the data sets of different analytical techniques. As model data sets, we will simulate data with specific properties and use existing data sets of different foods. Therefore, in addition to the generation of crucial methodological insights about the fusion of different molecular and elemental fingerprints, this project will also mark an important step towards improving food authentication for the revelation of food fraud.

• DFG Programme: Research Grants