Prof. Olga Kalinina, Helmholtz Institute for Pharmaceutical Research Saarland

Artificial intelligence has proven to be a useful tool in many areas of biomedical research, but its applications in natural product research are still scarce. The term natural products usually refers to chemical compounds produced by a living organism. Of particular interest are secondary metabolite natural products that are not directly involved in the organism’s growth, development or replication. They are often used to attack other species in the same ecological niche or for chemical communication, which makes them an obvious source of lead molecules for drug discovery.

In my talk, I will start with presenting a unique collection of organisms which harbor a great and underexplored potential for biosynthesis of natural products from the phylum Myxococcota, demonstrate some first applications of AI in this field, and then move on to discussing some more general problems of bioinformatics and machine learning such as predicting genotype-to-phenotype relationships. Another important topic that I will touch upon is minimizing data leakage while training AI models to avoid overinflation of reported performance and ensure models’ generalizability and applicability to out-of-distribution data.