Auditing Protein-Protein Interaction Signals with Sparse Autoencoder Fingerprints
- 1Shanghai Jiao Tong University
Abstract
Protein language models have become a dominant foundation for sequence-based protein-protein interaction (PPI) prediction, but their generalization remains limited under stringent evaluation, and benchmark accuracy alone cannot reveal whether a PPI predictor learns partner-specific biological signals or exploits contextual shortcuts. Here we introduce AuditPPI, an interpretable framework that transforms sparse-autoencoder (SAE) features from a frozen protein language model into order-invariant pair fingerprints, recasting PPI prediction as an auditable tabular-learning problem. AuditPPI achieves competitive predictive performance across diverse PPI benchmarks while enabling multiscale auditing of the information supporting its predictions. Protein-level audits show that partner-independent participation signals remain substantial in conventional benchmarks, including protein-disjoint benchmark, but are largely insufficient for pair-level discrimination in topology- and degree-controlled benchmark. Pair-level analyses further show that removing protein identity overlap does not eliminate contextual structure: protein-disjoint benchmarking retains strong subcellular co-localization signals and benchmark-dependent feature matching captured by SAE co-activation and absolute-difference. Structural analyses, however, provide little evidence that these predictive signals are specifically grounded in PPI interfaces: feature enrichment at interfaces for protein-disjoint benchmark was no longer detectable after controlling for surface exposure, and contact-specific enrichment was observed for only a small fraction of testable SAE feature pairs. AuditPPI therefore separates predictive success from mechanistic fidelity and provides a practical framework for evaluating whether sequence-based PPI predictors capture partner-specific biological signals instead of contextual shortcuts.
Related
- Are You Learning Biological Signal or Shortcuts? Auditing and Mitigating Bias in Protein-Protein Interaction Datasets
- MIMOSA: Guiding De Novo Antibody Design with Natural Interaction Fingerprints
- CUBE: Multimodal Representation Learning Reveals Biological Structure Across Histomorphology, Spatial Protein Phenotypes, and Transcriptome-Associated Signals
- TSNR: Protein-Specific Signals with Broad Molecular Context for Spatial Protein Prediction
- A non-canonical redox family protein regulates bacterial interaction machineries
BibTeX
@misc{zhu2026auditing,
title = {{Auditing Protein-Protein Interaction Signals with Sparse Autoencoder Fingerprints}},
author = {W. Zhu and S. Wang and X. Liu and Y. Xue and H.-B. Shen and B. Chen and X. Pan},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.27.754758},
url = {https://www.biorxiv.org/content/10.64898/2026.09.27.754758v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Auditing Protein-Protein Interaction Signals with Sparse Autoencoder Fingerprints
AU - W. Zhu
AU - S. Wang
AU - X. Liu
AU - Y. Xue
AU - H.-B. Shen
AU - B. Chen
AU - X. Pan
PY - 2026
DA - 2026/09/27
PB - bioRxiv
DO - 10.64898/2026.09.27.754758
UR - https://www.biorxiv.org/content/10.64898/2026.09.27.754758v1
AB - Protein language models have become a dominant foundation for sequence-based protein-protein interaction (PPI) prediction, but their generalization remains limited under stringent evaluation, and benchmark accuracy alone cannot reveal whether a PPI predictor learns partner-specific biological signals or exploits contextual shortcuts. Here we introduce AuditPPI, an interpretable framework that transforms sparse-autoencoder (SAE) features from a frozen protein language model into order-invariant pair fingerprints, recasting PPI prediction as an auditable tabular-learning problem. AuditPPI achieves competitive predictive performance across diverse PPI benchmarks while enabling multiscale auditing of the information supporting its predictions. Protein-level audits show that partner-independent participation signals remain substantial in conventional benchmarks, including protein-disjoint benchmark, but are largely insufficient for pair-level discrimination in topology- and degree-controlled benchmark. Pair-level analyses further show that removing protein identity overlap does not eliminate contextual structure: protein-disjoint benchmarking retains strong subcellular co-localization signals and benchmark-dependent feature matching captured by SAE co-activation and absolute-difference. Structural analyses, however, provide little evidence that these predictive signals are specifically grounded in PPI interfaces: feature enrichment at interfaces for protein-disjoint benchmark was no longer detectable after controlling for surface exposure, and contact-specific enrichment was observed for only a small fraction of testable SAE feature pairs. AuditPPI therefore separates predictive success from mechanistic fidelity and provides a practical framework for evaluating whether sequence-based PPI predictors capture partner-specific biological signals instead of contextual shortcuts.
ER -