Ontology Extension by Automated Learning and Reasoning from Structured Entities
Overview
This DFG- and SNF-funded project is a collaboration with the group of Prof. Dr. Till Mossakowski and Dr. Fabian Neuhaus. We have previously added an ontology training step to a deep learning model to improve the performance in the domain of toxicity prediction. Now, as one part of this project, we are extending this approach to improve predictive performance of a variety of other chemical properties, starting with solubility. Solubility is an important chemical property, especially in the field of drug design, where a certain solubility is vital for absorption.
Methods
For this project we are working with Transformer methods as well as Ontologies. Our application are chemical entities, we therefore use the Chebi Ontology.
Results/Publication
Our results so far suggest an improvement in predictive performance when adding the ontology pre-training, for details for the toxicity problem see this publication. Preliminary results on solubility confirm this trend. Stay tuned for updates!
People involved
As well as our collaborators in Germany: Simon Flügel, Martin Glauer, Fabian Neuhaus, Till Mossakowski