ppIRIS: deep learning for proteome-wide prediction of bacterial protein-protein interactions

Abstract Protein-protein interactions (PPIs) are central to cellular processes and host-pathogen dynamics, yet bacterial interactomes remain poorly mapped, especially for extracellular effectors and cross-species interactions. Experimental approaches provide only partial coverage, while existing computational methods often lack generalizability or are too resource-intensive for proteome-scale application. Here, we introduce ppIRIS (protein-protein Interaction Regression via Iterative Siamese networks), a lightweight deep learning model that integrates evolutionary and structural embeddings to predict PPIs directly from sequence. Trained on curated bacterial datasets, ppIRIS achieves state-of-the-art accuracy across benchmarks while enabling proteome-wide screening in minutes. Applied to Group A Streptococcus (GAS), ppIRIS revealed functional clusters linked to virulence pathways, including nutrient transport, stress response, and metal scavenging. For host-pathogen predictions, ppIRIS recovered 56.2% of known GAS-human plasma interactions with enrichment in complement, coagulation, and protease inhibition pathways. Experimental validation confirmed novel predictions, demonstrating the applicability of ppIRIS for systematic discovery of bacterial and cross-species PPIs. The software is freely available at github.com/lupiochi/ppIRIS. SuppZenodo_intra.zip contains the AlphaFold3 output and structures for complexes reported in the section "ppIRIS predicts proteome-scale PPIs in minutes" and Figure 3 of the main manuscript and Supplementary Figure S5. SuppZenodo_apms.zip contains the AlphaFold3 output and structures for the 63 'true positive' complexes reported in the section "Cross-species predictions between GAS M1 and human plasma proteins" and Figure 4 of the main manuscript.