An interpretable deep learning framework for classifying neuronal morphologies using topology and graph neural networks

Neuronal shape determines how neurons process and integrate information, yet a consistent and objective classification of neuronal morphologies remains elusive. Current approaches rely heavily on subjective expert views or on predefined features, limiting reproducibility and interpretability. Here, we present an interpretable deep learning framework that unifies topological data analysis, graph neural networks, and traditional morphometrics to classify neuronal morphologies objectively and transparently.

Our framework compares complementary mathematical representations of neurons to capture geometric, topological, and graph-structural information. Then it benchmarks their performance against expert-labeled datasets. We show that topology- and graph-based models achieve accuracies comparable to human experts, revealing that both global branching invariants and local connectivity patterns are essential to define morphological cell types.

Using explainable artificial intelligence methods, we identify structural features driving each classification decision, bridging computational and neuroanatomical interpretations. This open source and reproducible approach provides a foundation for scalable, interpretable and biologically meaningful neuronal taxonomy, enabling consistent comparisons between data sets and species. Kathryn Hess for her useful contributions to the discussions on topological descriptors, Prof Pierre Vandergheynst for the useful feedback and the suggestion of the deepwalk model for comparison, and Jan Krepl for his useful contributions to the machine learning techniques.

We thank LNMC for the reconstructions and the experimental work (Data collectors: Rodrigo de Campos Perin, Shruti Muralidhar, Thomas Berger) that made this study possible. This study was supported by funding to the Blue Brain Project, a research center of the École polytechnique fédérale de Lausanne (EPFL), from the Swiss government’s ETH Board of the Swiss Federal Institutes of Technology. L.K. was supported by the Medical Research Council, UKRI (MR/Z504804/1).

For the purpose of open access, the author has applied a CC-BY public copyright licence to any author accepted manuscript version arising from this submission. Department of Mathematics, University of Oxford, Oxford, UK Blue Brain Project, École Polytechnique Fédérale de Lausanne (EPFL), Campus Biotech, 1202, Geneva, Switzerland Lida Kanari, Stanislav Schmidt, Francesco Casalegno, Emilie Delattre, Jelena Banjac Lukic, Ying Shi, Felix Schürmann & Henry Markram Laboratory of Neuroepigenetics, Department of Health Sciences and Technology of the ETH Zürich, Brain Research Institute, Institute for Neuroscience, Medical Faculty, University of Zürich, Zurich, Switzerland Laboratory of Neural Microcircuitry, École Polytechnique Fédérale de Lausanne (EPFL), 1015, Lausanne, Switzerland Signal Processing Laboratory, École Polytechnique Fédérale de Lausanne (EPFL), 1015, Lausanne, Switzerland The authors declare no competing interests. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.