From viral histories to protein structures: new directions for understanding virus evolution

Dr Mahan Ghafari discusses new ways of thinking about old problems, and how emerging technologies can expand our understanding of viruses.

Viruses evolve at remarkable speed. During an outbreak, their genomes can change over weeks and months, leaving behind genetic traces that help us reconstruct where they came from, how they spread, and whether they are adapting to new hosts or immune landscapes. But how can we use genomic data to understand how fast viruses evolve, and what shapes their path, especially for viruses which have the potential to cause a pandemic?

Two recent review articles led by our group (Översti et al. and Lytras et al.) reflect where this work is now heading. Together, they ask how we can move beyond reading virus genomes as strings of letters on a phylogenetic tree, and instead place those genetic changes in their biological context: the three-dimensional structures of virus proteins, the immune responses they encounter, and the statistical and new machine-learning tools that are beginning to connect sequence, structure, and function at scale.

From sites to structure to serology

For viruses such as influenza and SARS-CoV-2, evading a host’s immune response plays an important role in which variants succeed. The antigens (molecules which trigger immune response) of these viruses are therefore evolving to avoid detection.

Modern genomic surveillance capabilities have transformed our ability to monitor virus evolution. We can now track mutations at individual sites in viral genomes and see how they appear and spread across populations. But not all mutations matter in the same way. A change on the exposed surface of a viral protein, close to an antibody-binding site where the host immune system fights off the virus, may have very different consequences from a change buried deep inside the protein, where the structure is tightly constrained.

This is why structure matters. Viral proteins are not flat sequences: they fold into complex three-dimensional shapes. Mutations can alter those shapes and affect protein stability, change receptor binding sites, add or remove sugar molecules that help conceal the virus from antibodies, or help a virus escape recognition by antibodies. Serology, the study of how antibodies recognise viruses, adds another layer by showing whether a new variant is still recognised by immunity generated through previous infection or vaccination.

These layers need to be linked up: sites, structure, and serology. The aim is not simply to explain why past variants succeeded, but to support more prospective assessment of future variants. If we can better understand which mutations are structurally plausible, immunologically important, and likely to affect viral fitness, we may be better placed to understand the evolutionary trajectory of viruses, inform vaccine updates, therapeutic design, and public health risk assessment.

Looking deeper into viral evolution with AI

We can also look at a different timescale: the deep evolutionary history of viruses.

Over short timescales, viral genomes often contain enough similarity for us to reconstruct evolutionary relationships using genomic sequence data. But over much longer periods, especially for RNA viruses, those signals can disappear. Viral sequences change so much that distant relatives may no longer look related at all. This has long made it difficult to study the ancient origins of viral families, classify highly divergent viruses discovered through metagenomics, or infer the functions of unfamiliar viral proteins.

Protein structure offers a way through this problem. Structure is often more conserved than sequence because proteins must retain shapes that allow them to function. Historically, however, experimentally solving protein structures was slow and expensive, meaning that structural information was available for only a small fraction of viral diversity.

AI-based structure prediction has changed that landscape dramatically over the past few years. Tools such as AlphaFold, ESMFold, RoseTTAFold, and related protein language models can now generate structural predictions at a scale that was unimaginable only a few years ago. This opens up new possibilities for virology: identifying distant evolutionary relationships, exploring “viral dark matter” – viral genes which we don’t know the functions of – in metagenomic datasets, improving virus taxonomy, and predicting the function of divergent proteins.

At the same time, we do need to be cautious. Predicted structures are powerful, but they are still models; they come with uncertainty and using them for evolutionary inference requires careful validation. The opportunity is not to replace classical virology or phylogenetics, but to extend them.

A broader direction for the group

Together, these reviews mark an important direction for our group. We are interested in how viruses evolve across timescales: from the emergence of a new outbreak lineage to the deep evolutionary origins of virus families. Increasingly, we want to connect that evolutionary history to mechanism. What do mutations actually do? Which changes are tolerated by protein structure? Which alter immune recognition? Which combinations of mutations are likely to matter for viral fitness?

Answering these questions will require a combination of approaches: phylogenetics, structural biology, antigenic data, experimental measurements, and machine learning. It will also require close collaboration across disciplines. No single dataset can fully explain viral evolution; each provides a different view of the same underlying process.

If we want to effectively prepare for a pandemic, this matters – the next important variant or emerging virus will not arrive with a complete explanation attached. We will need tools that can interpret limited and imperfect data quickly and responsibly. By bringing together virus genomics, phylogenetics, protein structure, and machine learning, we hope to build a more mechanistic understanding of pathogen evolution, one that helps us reconstruct where viruses have been, and perhaps anticipate where they may be able to go next.

To read more about serology and antigenic evolution (Översti et al. 2026), visit: https://journals.asm.org/doi/10.1128/jvi.01687-25

To read more about deep evolutionary history and protein structure (Lytras et al. 2026), visit: https://www.annualreviews.org/content/journals/10.1146/annurev-virology-100424-122154