EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Ontology pre-training improves machine learning-based predictions for metabolites

Charlotte Tumescheit iD, Martin Glauer, Simon Flügel iD, Martin Larralde iD, Fabian Neuhaus iD, Till Mossakowski, Janna Hastings iD

PLOS Computational Biology · 17.11.2026 · 22 (11) · e1014925

DOI: 10.1371/journal.pcbi.1014925 ↗

Kurz gefasst

Fachartikel in der Zeitschrift PLOS Computational Biology, online erschienen am 17.11.2026. Verfasst von Charlotte Tumescheit und 6 weiteren Autor:innen. Beteiligte Einrichtung: Idiap Research Institute, Martigny, Switzerland. Die Arbeit nennt 63 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung, Deutsche Forschungsgemeinschaft.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Deep learning has recently led to multiple breakthroughs in longstanding biological problems. Recent advances have shown that integration of expert knowledge can improve deep learning model predictive performance in complex domains such as biology. Bio-ontologies offer a rich source of curated biological knowledge that can be harnessed to this end. Here, we describe an intuitive and generalisable approach to embed the knowledge contained in a classification hierarchy derived from a bio-ontology into a deep learning model as an intermediate training step between general-purpose pre-training and task-specific fine-tuning: a process that we call’ontology pre-training’. We show that this approach leads to an improvement in predictive performance and a reduction in training time for a broad range of prediction tasks related to metabolite functions, activities and properties in living systems, using a range of datasets derived from MoleculeNet. We see the biggest improvement for regression tasks, e.g., prediction of lipophilicity and aqueous solubility of molecules, and a robust improvement for most classification tasks. Our approach can be adapted for a wide range of knowledge sources, models and prediction tasks.

Autor:innen und Institutionen

  • Idiap Research Institute, Martigny, Switzerland.
  • Swiss Institute of Bioinformatics, Lausanne, Switzerland.
  • Institute for Intelligent Cooperating Systems, Faculty of Computer Science, Otto-von-Guericke University Magdeburg, Magdeburg, Germany.
  • Institute of Computer Science, Osnabrück University, Osnabrück, Germany.
  • Leiden University Center for Infectious Diseases (LUCID), Leiden University Medical Center, Leiden, Netherlands.
  • European Molecular Biology Laboratory, Heidelberg, Germany.

ORCID-Profile: Charlotte Tumescheit, Simon Flügel, Martin Larralde, Fabian Neuhaus, Janna Hastings

Förderung

  • Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung
  • Deutsche Forschungsgemeinschaft

Zitieren

APA

Tumescheit, C., Glauer, M., Flügel, S., Larralde, M., Neuhaus, F., Mossakowski, T., & Hastings, J. (2026). Ontology pre-training improves machine learning-based predictions for metabolites. PLOS Computational Biology, 22(11), e1014925. https://doi.org/10.1371/journal.pcbi.1014925

Vancouver

Tumescheit C, Glauer M, Flügel S, Larralde M, Neuhaus F, Mossakowski T, et al. Ontology pre-training improves machine learning-based predictions for metabolites. PLOS Computational Biology. 2026;22(11):e1014925. doi:10.1371/journal.pcbi.1014925

BibTeX
@article{Tumescheit2026,
  author = {Tumescheit, Charlotte and Glauer, Martin and Flügel, Simon and Larralde, Martin and Neuhaus, Fabian and Mossakowski, Till and Hastings, Janna},
  title = {Ontology pre-training improves machine learning-based predictions for metabolites},
  journal = {PLOS Computational Biology},
  year = {2026},
  volume = {22},
  number = {11},
  pages = {e1014925},
  doi = {10.1371/journal.pcbi.1014925}
}
Kostenlos starten →