Poster De Conférence Année : 2024

Leveraging ontology structure for machine learning in knowledge graphs

Mise en valeur des ontologies pour l'apprentissage automatique dans les graphes de connaissances

Résumé

Understanding living organisms involves discovering connections between biological entities such as genes, proteins, or phenotypes. This knowledge can be represented as Biological Knowledge Graphs (BKGs) and forms a substantial body of data. BKGs often incorporate multiple ontologies [1-2], hierarchical structures that classify concepts for describing entities like genes or proteins. Leveraging these structures, machine learning methods have been applied successfully to predict new relationships between entities within BKGs [3]. However, we believe there is room for improvement. Indeed, integration of ontologies enhances not only the information content of the graph but also enables the use of specialized mathematical tools for analyzing graph structures efficiently. Specifically, sophisticated similarity indices between nodes can be calculated [3-4]. This work focuses on harnessing ontology-driven mathematical tools to enhance the performance of machine learning methods on BKGs. Through experiments conducted on a real-world dataset focusing on Japanese rice (Oryza sativa japonica) with both geometric and deep learning methods, we present strategies for harnessing ontological information within graphs. These strategies aim to improve the prediction of relationships between biological entities, offering valuable insights for downstream experimental exploration. The different strategies have in common the use of similarity measures to accurately quantify models mistakes during training. To date, we obtained slightly improved results compared to our benchmark. The work is ongoing and both new strategies and improvements of existing ones are being considered. References : 1.Caufield H, Putman T, Schaper K, Unni D, Hegde H, Callahan T, et al. KG-Hub - Building and Exchanging Biological Knowledge Graph [Internet]. 2023 [cited 2024 May 13]. Available from: https://europepmc.org/article/PPR/PPR621811 2Morris JH, Soman K, Akbas RE, Zhou X, Smith B, Meng EC, et al. The scalable precision medicine open knowledge engine (SPOKE): a massive knowledge graph of biomedical information. Lu Z, editor. Bioinformatics [Internet]. 2023 Feb 3 [cited 2024 May 13]. Available from: https://academic.oup.com/bioinformatics/article/doi/10.1093/bioinformatics/btad080/7033465 3. Kulmanov M, Smaili FZ, Gao X, Hoehndorf R. Semantic similarity and machine learning with ontologies. Briefings in Bioinformatics [Internet]. 2021 Jul 20 [cited 2024 May 13]. Available from: https://academic.oup.com/bib/article/doi/10.1093/bib/bbaa199/5922325 4.Kamran AB, Naveed H. GOntoSim: a semantic similarity measure based on LCA and common descendants. Sci Rep [Internet]. 2022 Mar 9 [cited 2024 May 13];12(1):3818. Available from: https://www.nature.com/articles/s41598-022-07624-3

Fichier principal
Vignette du fichier
JOBIM2024-11.pdf (867.62 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-04636636 , version 1 (05-07-2024)

Licence

Identifiants

  • HAL Id : hal-04636636 , version 1

Citer

Elliot Hugo Butz, Jérôme Azé, Konstantin Todorov, Antoine Toffano, Pierre Larmande. Leveraging ontology structure for machine learning in knowledge graphs. JOBIM 2024 - 24es Journées Ouvertes en Biologie, Informatique et Mathématiques, Jun 2024, Toulouse, France. ⟨hal-04636636⟩
262 Consultations
194 Téléchargements

Partager

  • More