GraPHFormer: A Multimodal Graph Persistent Homology Transformer for the Analysis of Neuroscience Morphologies

TL;DR

GraPHFormer is a novel multimodal transformer that integrates topological and geometric features of neuronal morphologies using contrastive learning, achieving state-of-the-art results across multiple neuroscience benchmarks.

Contribution

It introduces a multimodal architecture combining topological images and graph attributes with contrastive learning for neuronal morphology analysis.

Findings

Achieves state-of-the-art performance on five neuroscience benchmarks.

Effectively discriminates glial morphologies across regions and species.

Detects developmental and degenerative signatures in neuronal data.

Abstract

Neuronal morphology encodes critical information about circuit function, development, and disease, yet current methods analyze topology or graph structure in isolation. We introduce GraPHFormer, a multimodal architecture that unifies these complementary views through CLIP-style contrastive learning. Our vision branch processes a novel three-channel persistence image encoding unweighted, persistence-weighted, and radius-weighted topological densities via DINOv2-ViT-S. In parallel, a TreeLSTM encoder captures geometric and radial attributes from skeleton graphs. Both project to a shared embedding space trained with symmetric InfoNCE loss, augmented by persistence-space transformations that preserve topological semantics. Evaluated on six benchmarks (BIL-6, ACT-4, JML-4, N7, M1-Cell, M1-REG) spanning self-supervised and supervised settings, GraPHFormer achieves state-of-the-art performance on five benchmarks, significantly outperforming topology-only, graph-only, and morphometrics baselines. We demonstrate practical utility by discriminating glial morphologies across cortical regions and species, and detecting signatures of developmental and degenerative processes. Code: https://github.com/Uzshah/GraPHFormer