From Protein Interactions to Functional Annotation: Graph Alignment in Herpes

Sequence alignment forms the basis of many methods for functional annotation by phylogenetic comparison, but becomes unreliable in the `twilight’ regions of high sequence divergence and short gene length. Here we perform a cross-species comparison of two herpesviruses, VZV and KSHV, with a hybrid method called graph alignment. The method is based jointly on the similarity of protein interaction networks and on sequence similarity. In our alignment, we find open reading frames for which interaction similarity concurs with a low level of sequence similarity, thus confirming the evolutionary relationship. In addition, we find high levels of interaction similarity between open reading frames without any detectable sequence similarity. The functional predictions derived from this alignment are consistent with genomic position and gene expression data.

💡 Research Summary

The paper addresses a fundamental limitation of traditional sequence‑based annotation methods, which become unreliable in regions of high divergence and short gene length – the so‑called “twilight zone.” To overcome this, the authors develop a hybrid approach called graph alignment that simultaneously exploits protein‑protein interaction (PPI) network similarity and sequence similarity. Using two medically important herpesviruses, varicella‑zoster virus (VZV) and Kaposi’s sarcoma‑associated herpesvirus (KSHV), they construct separate PPI graphs for each virus and compute a pairwise sequence similarity matrix for all open reading frames (ORFs).

The core of the method is a composite objective function that combines (i) a weighted score derived from conventional BLAST‑like sequence alignments and (ii) a structural similarity term that rewards preservation of edges (interactions) between matched vertices. The authors formulate the alignment problem as a binary integer linear program and solve it with a flow‑based optimization algorithm, thereby finding a mapping of VZV ORFs onto KSHV ORFs that maximizes overall network concordance while respecting sequence evidence.

The resulting alignment falls into three distinct categories. The first comprises ORF pairs with high sequence and interaction similarity, confirming known orthologous relationships and serving as a sanity check. The second includes pairs with low sequence similarity but strong interaction similarity; these cases reveal evolutionary connections that would be missed by sequence‑only methods. Many of these pairs belong to conserved functional modules such as DNA replication, capsid assembly, or immune evasion, suggesting that interaction topology can preserve functional signals even when primary sequence has diverged.

The most striking finding is the third category: ORF pairs that show virtually no detectable sequence similarity yet exhibit a high degree of interaction similarity. These matches often involve proteins that occupy central positions in the respective PPI networks, implying that they are part of essential viral processes. The authors validate these predictions by cross‑referencing genomic location and RNA‑seq expression profiles. Matched ORFs tend to be co‑located in syntenic blocks and display synchronized expression peaks during the viral life cycle, providing independent evidence that the graph‑based matches reflect biologically meaningful relationships.

Beyond the specific VZV–KSHV comparison, the study demonstrates several broader implications. First, it proves that network‑level conservation can serve as a robust proxy for functional similarity when sequence information is insufficient. Second, the graph alignment framework is generic and can be extended to other virus families, bacterial systems, or even eukaryotic interactomes, provided that reliable interaction data are available. Third, the approach opens avenues for integrating additional data types—such as temporal interaction dynamics, host‑pathogen interaction maps, or structural domain information—into a unified alignment model, potentially increasing annotation accuracy further.

In conclusion, the authors present a compelling case that hybrid graph alignment, which jointly leverages sequence and interaction data, can uncover hidden evolutionary relationships and generate high‑confidence functional annotations for highly divergent viral genes. This methodology enriches the toolkit for viral genomics, supports the discovery of novel drug targets, and offers a scalable strategy for systematic functional annotation across diverse biological systems.