ISMB/ECCB 2023 Meeting

Ontology-based Interpretability for Large Predictive Models

Janna Hastings, University of Zurich

Abstract:

Bio-ontologies include consensus hierarchical classifications of entities in their domain and are a key resource for biomedical data-driven discovery research. While the potential uses of large-scale machine learning models for biomedical discoveries are growing, there are well-known challenges when such models operate as 'black boxes' that cannot be inspected: the model may learn spurious associations that lack generalisability, and the opacity may hinder the utility of the prediction in discovery research where novel findings need to be understood. Interpretability is provided for such models either extrinsically by determining feature importance for predictions, or intrinsically through inspecting model internal parameters such as Transformer attention weights. Ontologies are able to support the provision of natural explanations at different hierarchical levels, but are not accessed by most current approaches that provide interpretability for machine learning models. Through case studies in metabolism and in physical activity, I will show how ontologies can be used to supplement and extend existing approaches that enable interpretability of large machine learning models. 

Biography: 

Janna Hastings has a PhD in Computational Biology from the University of Cambridge (2019) and Masters in both Computer Science (2011) and Philosophy (2012). She has been actively developing bio-ontologies for more than a decade and was the project lead of the ChEBI ontology for a number of years.  She is currently Assistant Professor of Medical Knowledge and Decision Support at the Institute for Implementation Science in Health Care, Faculty of Medicine, University of Zurich, and Vice-Director of the School of Medicine at the University of St. Gallen. Her research focuses on how information systems can be better adapted to suit the needs of clinicians, with a special focus on how the newer, powerful approaches to artificial intelligence can be made more interpretable through using ontologies.

Ontology Alignment for Life Sciences

Ernesto Jimenez-Ruiz, City, University of London

Abstract:

The semantic web and life sciences research communities have extensively investigated the problem of defining correspondences between independently developed ontologies, which is usually referred to as the ontology alignment problem. Resulting from this effort are the growing number of ontology matching systems in development and the large mapping repositories that have been created. In particular, the ontology matching community has also been running an annual evaluation campaign (the OAEI) to benchmark ontology alignment systems over different matching tasks. Despite some joint efforts, the OAEI could be better aligned to real-world challenges from life sciences to fulfill not only the objective of improving automated ontology alignment systems but also providing useful outcomes that could be (re)used in practice. 

Biography: 

Ernesto Jimenez Ruiz is a Lecturer in Artificial Intelligence at City, University of London affiliated to the Research Centers for Machine Learning and Artificial Intelligence. He is also a researcher in the Centre for Scalable Data Access (SIRIUS) at the University of Oslo, Norway. He previously held a Senior Research Associate position at The Alan Turing Institute in London (UK) and a Research Assistant position at the University of Oxford. His home university (Universitat Jaume I, Castellon, Spain) awarded a “Premio extraordinario de doctorado” (roughly translated as a Extraordinary Doctoral Award) to his doctoral thesis (Engineering category 2010-2011). His research has covered several areas, including bio-medical information processing and integration, ontology reuse, ontology versioning and evolution, ontology alignment. His current research interests focus on the application of Semantic Technology to Data Science workflows and the combination of Knowledge Representation and Machine Learning techniques. Ernesto co-organizes the annual workshop on Ontology Matching together with the associated OAEI evaluation campaign. He is also co-organising the Semantic Web challenge on Tabular Data to Knowledge Graph Matching (SemTab) and the International Workshop on

Neural-Symbolic Learning and Reasoning.