ClusterGraph: a new tool for visualisation and compression of multidimensional data
GigaScience · 21.01.2026 · 15
DOI: 10.1093/gigascience/giag070 ↗Kurz gefasst
Fachartikel in der Zeitschrift GigaScience, online erschienen am 13.06.2026. Verfasst von Paweł Dłotko und 3 weiteren Autor:innen. Beteiligte Einrichtung: Center of Trustworthy AI for Life Sciences - International Research Agendas Programme, Warsaw University, 00-927, Warsaw, Poland. Hauptthemen laut MEDLINE: Data Compression, Software, Computational Biology. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Max Planck Society, National Science Centre (Poland), Polish Ministry of Science and Higher Education and the German Federal Ministry of Research, Technology, and Space.
Abstract (Originalsprache)
Background
Understanding the organization of high-dimensional data is of primary interest for many branches of applied sciences. It is typically achieved by applying dimensionality reduction techniques, which, while preserving local features, often miss the global structure of the dataset. Clustering techniques are another class of methods operating in the ambient space, grouping together similar points. However, unlike dimensionality reduction techniques, they do not provide information about the organization of the data.
Results
Leveraging ideas from Topological Data Analysis, we introduce ClusterGraph, an additional layer on the output of any clustering algorithm that represents clusters as vertices and inter-cluster relationships as weighted edges. This structure captures the large-scale organisation of a dataset without forcing it into a two-dimensional Euclidean embedding. The method comes with a built-in quality criterion, metric distortion, which quantifies how faithfully the graph reflects the intrinsic geometry of the data and provides a principled basis for pruning, model selection, and parameter tuning. ClusterGraph, possibly with an appropriate structure-preserving simplification, can be visualized and used in synergy with state-of-the-art exploratory data analysis techniques.
Conclusion
ClusterGraph is complementary to methods such as UMAP, t-SNE, and PHATE: by encoding inter-cluster geometry directly in graph form, it serves not only as a visualisation tool but also as a diagnostic layer that can validate, refine, or question conclusions drawn from low-dimensional embeddings.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Center of Trustworthy AI for Life Sciences - International Research Agendas Programme, Warsaw University, 00-927, Warsaw, Poland.
- Dioscuri Centre in Topological Data Analysis, Mathematical Institute, Polish Academy of Sciences, 00-656, Warsaw, Poland.
- Génie Informatique, Université de Technologie de Compiègne, 60200, Compiègne, France.
- School of Electronics, Electrical Engineering and Computer Science, Queens University of Belfast, BT9 5BN, Belfast, UK.
ORCID-Profile: Paweł Dłotko, Davide Gurnari, Mathis Hallier, Anna Jurek-Loughrey
Förderung
- Max Planck Society
- National Science Centre (Poland)
- Polish Ministry of Science and Higher Education and the German Federal Ministry of Research, Technology, and Space
- International Research Agendas programme of the Foundation for Polish Science
- European Union under the European Funds for Smart Economy 2021-2027 (FENG)
Zitieren
APA
Dłotko, P., Gurnari, D., Hallier, M., & Jurek-Loughrey, A. (2026). ClusterGraph: a new tool for visualisation and compression of multidimensional data. GigaScience, 15. https://doi.org/10.1093/gigascience/giag070
Vancouver
Dłotko P, Gurnari D, Hallier M, Jurek-Loughrey A. ClusterGraph: a new tool for visualisation and compression of multidimensional data. GigaScience. 2026;15. doi:10.1093/gigascience/giag070
BibTeX
@article{Dotko2026,
author = {Dłotko, Paweł and Gurnari, Davide and Hallier, Mathis and Jurek-Loughrey, Anna},
title = {ClusterGraph: a new tool for visualisation and compression of multidimensional data},
journal = {GigaScience},
year = {2026},
volume = {15},
doi = {10.1093/gigascience/giag070}
}
EUVIMED