Escaping the Phylogenetic Trap: Persistent Homology Enables Cross-Species Ion Channel Classification via Multimodal Graph Neural Networks
Abstract
Protein classifiers are highly sensitive to evaluation protocol: random train/test splits induce phylogenetic leakage, allowing models to exploit taxon-specific correlations and substantially overestimate out-of-distribution performance. This issue is particularly acute for venom compound classification by ion-channel target (Ca2+, K+, Na+), where reported accuracies of 80-90% do not reflect generalization to unseen taxa. We formulate cross-species prediction as a structured distribution shift in which taxonomic identity acts as a confounder, requiring representations that capture structural features invariant across phylogeny. Using a shared multimodal backbone that fuses ESM-2 sequence embeddings with a disulfide-aware graph attention network, we systematically compare five structural inductive biases (spectral, Riemannian, SE(3)-invariant, statistical-manifold, and persistent-homology representations) under a leave-one-taxon-out evaluation protocol on a benchmark of 1,373 ColabFold-predicted venom compounds spanning five taxa. The persistent-homology representation summarizes backbone geometry through topological descriptors that are robust to local structural perturbations, providing an inductive bias toward phylogenetically invariant learning. Under conventional i.i.d. splits, all models achieve similar accuracy (81-86%), masking differences in inductive bias. Under taxonomic shift, this parity collapses: a sequence-only baseline drops to approximately 30%, the remaining geometric priors cluster at 54-55%, while our persistent-homology model (PH-GNN) achieves 63.4%, outperforming the next-best method by over 8 percentage points. These results demonstrate that random-split evaluation obscures meaningful differences in representation quality and identify topological representations as an effective inductive bias for learning under structured biological distribution shift.
Related
- IonFusion: Multimodal Fusion for Hierarchical Classification and Functional Annotation of Ion Channels and Transporters
- When Clinical and Metabolomics Data Work Together: A Comparative Framework for Multimodal Disease Classification across Machine and Deep Learning.
- Identifiability of Large Phylogenetic Mixtures for Many Phylogenetic Model Structures.
- DNA Barcoding and Untargeted Metabolomics Reveal Phylogenetic and Chemotaxonomic Relationships Among 11 Ferula Species from Xinjiang, China
- Rad55-Rad57 and Srs2 regulate homology search onset, coordination, reach and inactivation
BibTeX
@misc{bybordi2026escaping,
title = {{Escaping the Phylogenetic Trap: Persistent Homology Enables Cross-Species Ion Channel Classification via Multimodal Graph Neural Networks}},
author = {Arezoo Bybordi and Favour Achimba and Mariam Gelashvili and Jessy Ramirez and Weigang Qiu and Mandë Holford and Anita Raja},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.30394},
eprint = {2610.30394},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.30394},
}
RIS
TY - UNPB
TI - Escaping the Phylogenetic Trap: Persistent Homology Enables Cross-Species Ion Channel Classification via Multimodal Graph Neural Networks
AU - Arezoo Bybordi
AU - Favour Achimba
AU - Mariam Gelashvili
AU - Jessy Ramirez
AU - Weigang Qiu
AU - Mandë Holford
AU - Anita Raja
PY - 2026
DA - 2026/10/21
PB - arXiv
DO - 10.48550/arxiv.2610.30394
UR - https://arxiv.org/abs/2610.30394
AB - Protein classifiers are highly sensitive to evaluation protocol: random train/test splits induce phylogenetic leakage, allowing models to exploit taxon-specific correlations and substantially overestimate out-of-distribution performance. This issue is particularly acute for venom compound classification by ion-channel target (Ca2+, K+, Na+), where reported accuracies of 80-90% do not reflect generalization to unseen taxa. We formulate cross-species prediction as a structured distribution shift in which taxonomic identity acts as a confounder, requiring representations that capture structural features invariant across phylogeny. Using a shared multimodal backbone that fuses ESM-2 sequence embeddings with a disulfide-aware graph attention network, we systematically compare five structural inductive biases (spectral, Riemannian, SE(3)-invariant, statistical-manifold, and persistent-homology representations) under a leave-one-taxon-out evaluation protocol on a benchmark of 1,373 ColabFold-predicted venom compounds spanning five taxa. The persistent-homology representation summarizes backbone geometry through topological descriptors that are robust to local structural perturbations, providing an inductive bias toward phylogenetically invariant learning. Under conventional i.i.d. splits, all models achieve similar accuracy (81-86%), masking differences in inductive bias. Under taxonomic shift, this parity collapses: a sequence-only baseline drops to approximately 30%, the remaining geometric priors cluster at 54-55%, while our persistent-homology model (PH-GNN) achieves 63.4%, outperforming the next-best method by over 8 percentage points. These results demonstrate that random-split evaluation obscures meaningful differences in representation quality and identify topological representations as an effective inductive bias for learning under structured biological distribution shift.
ER -