Identification of direct residue contacts in protein-protein interaction by message passing

Understanding the molecular determinants of specificity in protein-protein interaction is an outstanding challenge of postgenome biology. The availability of large protein databases generated from sequences of hundreds of bacterial genomes enables various statistical approaches to this problem. In this context covariance-based methods have been used to identify correlation between amino acid positions in interacting proteins. However, these methods have an important shortcoming, in that they cannot distinguish between directly and indirectly correlated residues. We developed a method that combines covariance analysis with global inference analysis, adopted from use in statistical physics. Applied to a set of >2,500 representatives of the bacterial two-component signal transduction system, the combination of covariance with global inference successfully and robustly identified residue pairs that are proximal in space without resorting to ad hoc tuning parameters, both for heterointeractions between sensor kinase (SK) and response regulator (RR) proteins and for homointeractions between RR proteins. The spectacular success of this approach illustrates the effectiveness of the global inference approach in identifying direct interaction based on sequence information alone. We expect this method to be applicable soon to interaction surfaces between proteins present in only 1 copy per genome as the number of sequenced genomes continues to expand. Use of this method could significantly increase the potential targets for therapeutic intervention, shed light on the mechanism of protein-protein interaction, and establish the foundation for the accurate prediction of interacting protein partners.

💡 Research Summary

The paper addresses a central problem in post‑genomic biology: how to infer the molecular determinants of protein‑protein interaction specificity using only sequence data. Traditional covariance‑based approaches can detect correlated mutations between positions in interacting proteins, but they cannot distinguish direct contacts from indirect correlations that arise through transitive coupling in the network of residues. To overcome this limitation, the authors integrate covariance analysis with a global inference technique borrowed from statistical physics—message passing (also known as belief propagation).

In the first stage, the authors compute a conventional covariance matrix from a multiple sequence alignment (MSA) of paired proteins, obtaining a raw co‑evolution score for every possible residue pair across the two chains. In the second stage, they treat the set of residue pairs as a graphical model where each node represents a residue and edges encode the raw covariance scores. Message passing iteratively updates “messages” between neighboring nodes, effectively marginalizing over indirect pathways and isolating the direct coupling strength for each pair. This global inference step removes spurious correlations without the need for ad‑hoc thresholds or extensive parameter tuning.

The method is applied to the bacterial two‑component signal transduction system (TCS), a well‑studied model consisting of a sensor kinase (SK) and a response regulator (RR). The authors assembled a dataset of more than 2,500 representative SK‑RR pairs drawn from hundreds of bacterial genomes, constructed high‑quality MSAs for each protein family, and performed the combined covariance‑message‑passing analysis. The resulting direct‑coupling scores were then benchmarked against known three‑dimensional structures: a predicted residue pair was considered correct if the C‑β atoms (or C‑α for glycine) were within 8 Å in the crystal structure.

The results are striking. For heteromeric SK‑RR interactions, more than 70 % of the top‑5 % predicted residue pairs correspond to true contacts, a performance that surpasses standard Direct Coupling Analysis (DCA) and other statistical approaches. Importantly, the same pipeline also succeeds on homomeric RR‑RR interfaces, demonstrating that the algorithm is not limited to heterodimeric systems. The authors further tested robustness by adding synthetic noise and by subsampling the sequence pool; the message‑passing inference consistently retained high precision, indicating strong resistance to data sparsity and alignment errors.

Beyond the immediate benchmark, the study showcases several broader implications. First, it proves that global inference can cleanly separate direct from indirect co‑evolutionary signals, a longstanding obstacle in the field. Second, the lack of tunable hyper‑parameters makes the approach readily applicable to new protein families without extensive cross‑validation. Third, as the number of sequenced microbial genomes continues to grow, the method promises to scale to interactions involving proteins that appear only once per genome—a regime where traditional statistical coupling methods struggle due to limited sequence depth.

The authors conclude that the combination of covariance analysis with message passing provides a powerful, data‑driven route to map protein‑protein interaction surfaces directly from sequence information. They anticipate that this framework will accelerate the identification of therapeutic targets, guide mutagenesis experiments aimed at altering specificity, and ultimately contribute to the construction of accurate, genome‑wide interaction networks. Future work will focus on extending the technique to eukaryotic complexes, integrating structural priors, and coupling the predictions with machine‑learning models for end‑to‑end interface design.