Abstract

Food supplements, as well as some foods themselves, can adversely affect the efficacy and safety of oncological drugs, often rather dramatically. This is due to insufficiently investigated biochemical interactions of the ingredients of these compounds. For a number of reasons, predicting and mitigating these possibly adverse interactions is currently difficult in clinical practice, but the problem could be solved in an innovative and scalable way using knowledge graphs derived from biomedical networks.

Networks have long been a popular tool for describing complex biological processes, where they model the interactions between biological entities and their effects on different biological systems. Biomedical networks can be converted into so-called knowledge graphs, which are formal representations of entities and semantic relationships between them. As a result, knowledge graphs can be used to develop relational learning models that are known to provide highly scalable and accurate predictions of novel associations between biomedical entities such as drugs, proteins, or diseases. In addition, knowledge graphs can naturally serve to develop applications of explainable artificial intelligence, which is a de facto necessity in a clinical context. Such models can then be successfully used, among other things, to develop an automated application for alerting physicians and/or patients to possible adverse interactions between common dietary supplements and drugs administered in cancer therapies.

In this project, we would primarily work on the basic research needed to address the interaction between dietary supplements and cancer treatments. However, we would also focus on evaluating our explainable predictions using data obtained from patient records, whether in a retrospective validation or a newly designed prospective study. In addition, we would deal with the possible commercialization of the research project outcomes (e.g., in the form of a spin-off company).