P2PSigLip logo
P2PSigLip
Manuscript And Resource Guide

P2PSigLip: Proteome-Scale Protein Interaction Prioritisation

P2PSigLip is designed as a proteome-scale candidate prioritisation resource. Its scores are primarily intended to rank interaction candidates within a query or candidate pool for downstream structural modelling and experimental validation, rather than serving as calibrated interaction probabilities.

Interpretation rule: read P2PSigLip output as query-level prioritisation for follow-up. It should not be interpreted as a calibrated interaction probability or as a universal decision threshold; use annotations, homologs, local network context, and experiments to decide which candidate pairs deserve deeper investigation.
Abstract Snapshot
Method Summary

P2PSigLip uses frozen ProtT5 embeddings and dual projection heads to turn pairwise PPI screening into reusable retrieval. Each protein is embedded once, and proteome-scale all-vs-all scoring becomes a matrix operation rather than a separate forward pass per pair.

Under strict homology-controlled evaluation, the method improves partner-ranking performance over strong sequence baselines and matches or exceeds structure-aware references on several external benchmark settings. The resource is designed for candidate prioritisation at proteome scale.

The web resource exposes precomputed multi-species prioritisation-score sets so users can search by gene or protein ID, retrieve ranked partners, inspect local neighbourhoods, and download results without rerunning the scoring pipeline locally.

Overview
At A Glance
Formal documentation layout
Species Coverage
13+
Cross-kingdom release spanning plants and metazoans, with the current website now extending presentation across plants, animals, and fungi.
Experimental Highlight
15 / 15
Tomato RIN-centred Y2H validation for P2PSigLip-prioritised candidates, compared with 3 / 12 for structure-prioritised candidates.
Recommended Use
Use P2PSigLip for proteome-scale first-pass prioritisation within each query context, then follow top candidates with structure-aware modelling and targeted experiments.
Citation
Citation And Resources
Huang Y, Tang JC, Jia YH, Hong YM, Liu FQ, Zhang Z. P2PSigLip: Proteome-scale protein-interaction prioritisation with sigmoid-aligned protein language-model embeddings.
Author List
Yao Huang†, JiCheng Tang†, YuHan Jia†, YuMin Hong, FangQi Liu and Zhao Zhang*
† These authors contributed equally to this work.
* Correspondence: [email protected]