Periodic Reporting for period 1 - Onto-mQoL (Extensions of Temporal Representations for Ontology-based and Holistic Reasoning in the Mobile Quality of Life Domain)
Période du rapport: 2021-09-01 au 2023-08-31
While temporal aspects are usually present in the health domain, modelling information evolving time/events in ontologies is a complex problem because temporal relations are ternary and cannot be directly handled by ontology languages. This problem happens because ontologies use Description Logic (DL) as a formal basis, which is a decidable fragment of the first-order logic that only uses unary (concepts) and binary (roles) predicates. Current temporal extensions that are compatible with the existing ontology resources (e.g. reasoners and DL formalism) are limited in aspects related to uncertain time, evolution of events, and relations between time concepts. This lack of expressive and standards-compatible temporal representations restricts the range of reasoning processes because several time aspects, usually embedded in the health and QoL domain, cannot be employed. The expressive temporal notion is important for healthcare knowledge representations because it allows a more comprehensive understanding of patient's health history and how daily behaviors affect their well-being. Moreover, this knowledge can facilitate accurate diagnosis and treatment planning and underpin various aspects of healthcare and behavior change decision-making. Thus, such technology can enable the delivery of better health services that benefit society in general.
This project aimed to design a conceptual framework for representation and reasoning about temporal aspects associated with an ontological engineering method to guide the use of this novel framework while converting static ontologies to their temporal versions. Thus, the project advances the state of the art as follows:
• Extension of temporal representations so they consider different time concepts (moments and intervals), time concept properties (precise and uncertain), time relations (interval-interval, interval-moment, and moment-moment), and time relation properties (qualitative and quantitative).
• Recent approaches for temporal reasoning are based on theories (e.g. Fuzzy Theory Sets and Temporal DLs) incompatible with the current Semantic Web standards (e.g. DL formalism and reasoners). Maintaining this compatibility is an important premise and contribution of this present project.
• The associated engineering method is an innovation proposed in this project since other approaches do not present pragmatic strategies to apply their contributions.
• Finally, this project demonstrates the importance of holistic representations in temporal QoL domains using experiments with existent datasets.
This project aimed to design a conceptual framework for representation and reasoning about temporal aspects associated with an ontological engineering method to guide the use of this novel framework while converting static ontologies to their temporal versions. Thus, the project advances the state of the art as follows:
• Extension of temporal representations so they consider different time concepts (moments and intervals), time concept properties (precise and uncertain), time relations (interval-interval, interval-moment, and moment-moment), and time relation properties (qualitative and quantitative).
• Recent approaches for temporal reasoning are based on theories (e.g. Fuzzy Theory Sets and Temporal DLs) incompatible with the current Semantic Web standards (e.g. DL formalism and reasoners). Maintaining this compatibility is an important premise and contribution of this present project.
• The associated engineering method is an innovation proposed in this project since other approaches do not present pragmatic strategies to apply their contributions.
• Finally, this project demonstrates the importance of holistic representations in temporal QoL domains using experiments with existent datasets.
The work carried out during these two years can be explained in three parts. The first part was related to specifying a temporal framework for temporal representations and the associated engineering process to create temporal ontologies or convert static ontologies to their temporal versions. The main scientific achievement was describing the semantics involving different temporal aspects, such as the uncertainty of temporal relations and classes. Moreover, this proposal followed the OWL standards, so our approach can be integrated into the main ontology frameworks and semantic reasoning engines. Thus, our main contribution was to provide ontological support for domains that require more complex temporal representations, such as uncertain time and time relations, considering both temporal intervals and moments. The approach's viability and generalizability were demonstrated through practical applications using case studies in different health domains, such as mental health, aging, and sleep quality issues.
The second part of our project employed new forms of reasoning (inductive reasoning), which receive descriptions modelled according to our ontological proposal as input. For this work, we considered the state-of-the-art inductive methods, represented by the deep learning transformers architectures. Our review on this topic (paper submitted for review) shows that their definition only supports sequential rather than temporal information (e.g. events duration and temporal distance). Thus, our main contribution was to formalise such limitations and propose initial research directions to cover these limitations.
The third part was related to the work conducted during the secondment. The proposal was to adapt and evaluate the algorithms developed within the Onto-mQoL project on EHR-based datasets provided by the Imperial College London (Computational Oncology Group, Department of Surgery & Cancer, Faculty of Medicine). These activities involved, for example, the specialisation of our ontology to cover special aspects of cancer disease and its dataset population (e.g. ageing and gender).
As main results, we had three scientific publications in high-impact journals and a conference paper published at the IEEE International Conference on Biomedical and Health Informatics. We still have three journal papers in the review process. If such papers are not accepted, they will be deposited in open-access repositories. All these papers have the EU funding reference.
The exploitation and dissemination of the project were mainly conducted by means of scientific and general public presentations at local, regional, and international events: University of Geneva Data Science Day 2022, Conference d Universitarire de Suisse Occidentale (CUSO) Winter School 2022, ISOQOL Annual Conference 2022, and IEEE-EMBS International Conference on Biomedical and Health Informatics. Moreover, the project had an initial web page for promoting its progress. Such a page had an interactive simulation where visitors could play with the temporal reasoning engine, a blog with project updates, and a video demonstrating an application (versions with subtitles in English and French). A second version of this web page (current version) is hosted on the university's main server (https://www.unige.ch/onto-mqol/) and it contains all the supplementary material indicated in the papers. A project summary is also hosted on the projects page of our laboratory (https://www.qualityoflifetechnologies.com/research-project/onto-mqol/).
The second part of our project employed new forms of reasoning (inductive reasoning), which receive descriptions modelled according to our ontological proposal as input. For this work, we considered the state-of-the-art inductive methods, represented by the deep learning transformers architectures. Our review on this topic (paper submitted for review) shows that their definition only supports sequential rather than temporal information (e.g. events duration and temporal distance). Thus, our main contribution was to formalise such limitations and propose initial research directions to cover these limitations.
The third part was related to the work conducted during the secondment. The proposal was to adapt and evaluate the algorithms developed within the Onto-mQoL project on EHR-based datasets provided by the Imperial College London (Computational Oncology Group, Department of Surgery & Cancer, Faculty of Medicine). These activities involved, for example, the specialisation of our ontology to cover special aspects of cancer disease and its dataset population (e.g. ageing and gender).
As main results, we had three scientific publications in high-impact journals and a conference paper published at the IEEE International Conference on Biomedical and Health Informatics. We still have three journal papers in the review process. If such papers are not accepted, they will be deposited in open-access repositories. All these papers have the EU funding reference.
The exploitation and dissemination of the project were mainly conducted by means of scientific and general public presentations at local, regional, and international events: University of Geneva Data Science Day 2022, Conference d Universitarire de Suisse Occidentale (CUSO) Winter School 2022, ISOQOL Annual Conference 2022, and IEEE-EMBS International Conference on Biomedical and Health Informatics. Moreover, the project had an initial web page for promoting its progress. Such a page had an interactive simulation where visitors could play with the temporal reasoning engine, a blog with project updates, and a video demonstrating an application (versions with subtitles in English and French). A second version of this web page (current version) is hosted on the university's main server (https://www.unige.ch/onto-mqol/) and it contains all the supplementary material indicated in the papers. A project summary is also hosted on the projects page of our laboratory (https://www.qualityoflifetechnologies.com/research-project/onto-mqol/).
This project focused on providing new health technologies that could benefit society. For example, it allows the representation of medical knowledge that involves temporal notions and uncertainty. Moreover, it can be integrated into the state-of-the-art strategies of inductive reasoning (e.g. deep learning architectures), creating a new family of neuro-symbolic explainability approaches for such reasoning. The case studies demonstrated this potential in domains such as quality of life (e.g. sleep quality) and cancer (the leading cause of death in 2020). The knowledge obtained during this project was used to specification two new projects. The first project uses the knowledge acquired on knowledge representation (temporal ontologies), deep learning transformers, and cancer domain to propose a knowledge-based system that supports the definition of cancer treatments, considering patients’ quality of life rather than only their rate of survival. We will collaborate with the European Organisation for Research and Treatment of Cancer (EORTC) on the newly funded project in that area (2024-2028). The second project was submitted to the Swiss National Science Foundation (SNSF). It uses the same technology to analyse migraine-related data and identify migraine attacks/episodes.