Summary
Classifying venom proteins by their ion channel targets often fails when tested on species not seen during training, because models pick up on species-specific signals rather than universal features. The authors show that using a mathematical tool called persistent homology, which captures shape features of proteins that don't change across species, helps models generalize better. Their method, combining protein sequence data with these topological descriptors, outperforms other approaches when tested on venom proteins from different species. This suggests new ways to better understand protein function across diverse organisms.
What this means in practice
- •For biotech developers: Develop cross-species protein classifiers that better generalize by incorporating topological features of protein structures.
- •For pharmaceutical data teams: Improve venom compound function prediction tools to better identify therapeutic targets across diverse species using topology-based models.
Authors
Arezoo Bybordi, Favour Achimba, Mariam Gelashvili, Jessy Ramirez, Weigang Qiu, Mandë Holford, Anita Raja
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.