Personalised Nutrition

The interest in healthy nutrition is very high both in everyday life and in science. With healthy nutrition we can not only reach our ideal weight, but also increase our performance, improve clinical pictures and age healthily. For a long time there was a trend towards making generally valid statements about the benefits of certain foods. In the meantime, however, it has become clear that the same foods can also have a different effect on different people.

At the interface between nutritional medicine and medical informatics, we research influencing factors on blood sugar in cooperation with Perfood GmbH / MillionFriends. In addition to detailed information on nutrition, we also record individual parameters of the test subjects (phenotype) and the intestinal microbiome. Taking into account knowledge from nutritional medicine, we analyze these large data sets using learning-based methods of pattern recognition in order to better understand the relationship between significant widespread diseases (e.g. obesity, diabetes mellitus type 2, migraine) and nutrition. We use complex neural networks and probabilistic pattern recognition methods.

However, complex machine learning methods generate decisions that can be checked for correctness but cannot be semantically reproduced (Black Box Algorithms). Therefore, an important goal of our research in the field of personalized nutrition is to introduce learning-based pattern recognition methods that, despite their high degree of complexity, make the data flow from input to output comprehensible (Explainable Machine Learning). For this purpose, we investigate, among other things, hierarchical architectures of deep neural networks. Only with explainable methods of Artificial Intelligence we will finally be able to make the connections between the mentioned diseases and nutrition understandable.