Viruses rank among the most abundant biological entities in the biosphere and display remarkable genetic diversity. Their ability to infect bacteria, archaea, and eukaryotes has established viral systems as a significant area of biological research. Despite this, the functions of a large proportion of proteins encoded by viral genomes remain uncharacterized, limiting a comprehensive understanding of the biological roles and evolutionary relationships of viruses. Methods commonly used to determine protein function rely on similarities between amino acid sequences; however, the rapid evolution of viral genomes leads to a progressive decline in sequence similarity over time, rendering sequence-based analyses alone insufficient. Protein structure (three-dimensional folding architecture) is known to be more conserved throughout evolution than sequence, making structure-based analyses a more reliable approach for elucidating relationships between proteins. Nevertheless, a substantial fraction of viral proteins remains unrepresented in existing structural databases, hindering comprehensive analysis of the viral protein space.
In this study conducted by Odai and colleagues, the aim was to examine the structural diversity of viral proteins on a large scale and to evaluate the contribution of structure-based analyses in cases where sequence-based methods prove inadequate. The study sought to predict the three-dimensional structures of viral proteins, systematically analyze structural similarities among them, and investigate whether conserved structural features exist across viruses infecting different hosts. An additional objective was to characterize the structural properties of functionally uncharacterized proteins within the viral protein space and to establish a foundation for drawing inferences about their potential functions.
To this end, viral protein sequences from the NCBI RefSeq database were comprehensively collected and clustered using MMseqs2 at a 30% identity threshold and a 0.01 coverage threshold, with representative sequences selected for each cluster. High-resolution three-dimensional structural predictions covering both monomer and homodimer forms were then generated for 27,279 representative proteins using the AlphaFold2 algorithm (a deep learning-based protein structure prediction algorithm), enabling assessment of potential interaction states. The resulting structural models were evaluated using pLDDT (predicted Local Distance Difference Test) scores and demonstrated high reliability with a mean score of 78. These structures were subsequently compared against experimentally determined structures in the Protein Data Bank (PDB), and Foldseek (TM-align mode) was employed to perform comparative analyses focused on identifying proteins that share structural similarity despite lacking detectable sequence similarity.
The study resulted in the construction of the Viral AlphaFold Database (VAD), made publicly accessible at data-sharing.atkinson-lab.com/vad/, representing a broad viral protein space encompassing both monomer and homodimer structures. Analyses revealed conserved protein folds across viruses infecting bacteria, archaea, and eukaryotes, demonstrating that viral proteins can retain certain structural features throughout evolution and that many viral proteins share structural similarity despite lacking sequence similarity, thereby underscoring the potential of structure-based approaches in functional annotation. A substantial proportion of viral proteins did not match any known function, a phenomenon described as “functional darkness,” indicating that large portions of the viral protein space remain unexplored. The study further reported the identification and experimental validation of a previously uncharacterized type II toxin–antitoxin system designated KreTA, demonstrating that viral proteins may harbor novel biological mechanisms.
The findings are proposed to contribute to the understanding of structural and evolutionary relationships among viral proteins, and to serve as a foundation for future structural, functional, and comparative analyses. The construction of VAD as an openly accessible resource represents a step toward systematically addressing the functional and structural gaps that currently limit our knowledge of the viral protein space.
Translated by: Melissa Öner
Editor: Elinsu Ak
Referans: Odai, R., Leemann, M., Al-Murad, T., Abdullah, M., Shyrokova, L., Tenson, T., Hauryliuk, V., Durairaj, J., Pereira, J., & Atkinson, G. C. (2025b). The Viral AlphaFold Database of monomers and homodimers reveals conserved protein folds in viruses of bacteria, archaea, and eukaryotes. Science Advances, 11(40), eadz8560. https://doi.org/10.1126/sciadv.adz8560
-Bioinfocodes Scientific News Service-
News articles prepared by our team members, reviewing and compiling scientific research published in journals with an impact factor greater than 20 (click here for the list).