Faculty of Computer Science

Research Group Theoretical Computer Science

Oberseminar: Heterogene formale Methoden

Abstract:

Chemical ontologies such as ChEBI contain tens of thousands of chemical entities that have manually been classified into classes based on shared features in the chemical structure. Organic chemistry is highly combinatorial and even the number of possible “small” molecular chemicals is in the billions; new chemicals are being invented all the time, e.g. in drug discovery. Thus, it would be very useful to have automatic methods to classify novel chemical entities into classes based on their chemical structures. In cheminformatics, chemical structures are encoded in graphs; chemical graphs are frequently used for machine learning to predict, e.g. chemical activities based on structural features. However, machine learning does not work very well for predicting ChEBI ontology classes. One possible reason is that the ontology classification is massively polyhierarchical (i.e. entities have many parents). In this presentation I will show the results of some experiments with machine learning for automatically classifying chemical entities in ChEBI.