J Neural Eng. 2026 Feb 9. doi: 10.1088/1741-2552/ae4381. Online ahead of print.
ABSTRACT
OBJECTIVES: Neuron morphology plays a vital role in defining cellular identity and function, offering valuable insights for cell type classification and neurological disorder diagnosis. However, two main challenges hinder progress: the difficulty of learning meaningful representations from complex, tree-like structures, and the high cost of expert annotation for large-scale datasets.
APPROACH: To address these challenges, we propose MorphSys, a self-supervised contrastive learning framework that complements a Branch-Aware module and a GNN-based module. We present a branch-level representation of neuron morphology by introducing an Inter-Branch Attention, which captures inter-branch relationships that are overlooked by conventional tree-graph models relying on node-level message passing. We further demonstrate the effectiveness and interpretability of branch-level knowledge in learning meaningful representations of neuron morphology. Meanwhile, our GNN-based module shows a robust ability for various GNN architectures in learning local features of neuron tree graph, where we draw from results that GatedGraphConv with SumPool yields the superior performance.
MAIN RESULTS: Comprehensive experiments on multiple benchmark datasets indicate that MorphSys consistently outperforms existing methods in neuron cell type classification. On the BIL dataset, MorphSys achieves the KNN-Acc of 83.99%, surpassing the previous state-of-the-art by 2.99%. Ablation study is conducted to verify the efficacy of several components of MorphSys, while an in-depth discussion is performed to identify powerful approaches for branch feature extraction.
SIGNIFICANCE: These results highlight that MorphSys serves an effective tool for the representation learning of neuron morphology and morphology-driven neuronal analysis.
PMID:41662773 | DOI:10.1088/1741-2552/ae4381
