Neuronal arbors and botanical trees are branching structures whose characteristics depend on both connectivity and three-dimensional geometry.

This topic explores how graph transformers can learn from these structures by combining information about individual branches with relationships across the entire tree. Possible directions include representations of branching hierarchy and spatial geometry, learning from unlabelled examples and synthetic data, and transferring models between datasets or biological domains. Developed models will be evaluated against morphology classification or dataset-dependent regression tasks.

Starting literature:

• MorphRep: Learning Meaningful Representation of Single-neuron Morphology via Large-scale Pre-training (2024)
• NeuroGT (2026)
• GraphGPS (2022)