Ontology - Supported Machine Learning and Decision Support in Biomedicine

Nowadays, ontologies and machine learning constitute two major technologies for domain-specific knowledge extraction which are actively used in knowledge-based systems of different kind including expert systems, decision support systems, knowledge discovery systems, etc. While the aim of these two technologies is the same - the extraction of useful knowledge - little is known about how the two sources of knowledge can be successfully integrated. Today the two technologies are used mainly separate; even though the knowledge extracted by the two is complementary and significant benefits can be obtained if the technologies were integrated. This problem is especially important for biomedicine where relevant data are often naturally complex having large dimensionality and including heterogeneous features, and where a large body of knowledge is available in the form of ontologies. In this paper we propose one approach for improving the performance of machine learning algorithms by integrating the knowledge provided by ontologies. The basic idea is to redefine the concept of similarity for complex heterogeneous data by incorporating available ontological knowledge, creating a bridge between the two technologies. Potential benefits and difficulties of this integration are discussed, two techniques for empirical evaluation and fine-tuning of feature ontologies are described, and an example from the field of paediatric cardiology is given.