Viral protein structures added to AlphaFold Database for pandemic preparedness

AI-predicted protein complex structures for more than 2,800 viruses are now openly available through the AlphaFold Database. This provides researchers worldwide with new molecular information to support infectious disease research and preparedness for future outbreaks. The new dataset prioritises viral families including those known to infect humans and covers viruses ranging from common human pathogens to emerging threats, such as mpox. By providing predicted three-dimensional structures of viral proteins and protein complexes, the resource can help researchers investigate how viral proteins function and interact and identify potential targets for diagnostics, therapeutics, and vaccines.

The expanded AlphaFold Database illustrates how artificial intelligence and open data can contribute to a more anticipatory approach to pandemic preparedness. Having structural information available before an outbreak provides researchers with a valuable starting point for studying unfamiliar pathogens and identifying potential targets for medical countermeasures. However, experimental work remains essential in these efforts, as predicted structures cannot anticipate virus behaviour or host-pathogen interactions, for example. For more information about the release, see the press release from EMBL-EBI.

The release is the result of the work of multiple groups from around the world. It includes protein complexes independently computed by Gemma Atkinson’s group at Lund University, bringing a significant Swedish contribution to this international resource. The Viral AlphaFold Database (VAD) Atkinson lab project was spearheaded by PhD student Roni Odai and was a collaboration with Vasili Hauryliuk’s group in Lund University, and groups in Basel, Switzerland and Tartu, Estonia. It was published last year in Science Advances. As well as human infecting viruses, VAD contains proteins from plant, animal and protist viruses, and - importantly for the future development of phage therapy to combat antibiotic resistant infections - bacteria-infecting viruses, bacteriophages.

Many viral proteins are poorly characterised, as virus sequences change so rapidly. Structural predictions provide another way of identifying similarities between proteins and can offer clues about their possible functions. By comparing structures, rather than relying only on protein sequences, their research has revealed evolutionary relationships and conserved protein folds that can be difficult to identify from sequence information alone. The Atkinson group is now building on their previous work by focusing more deeply on viral lineages that include human pathogens, and building an explorative pipeline to discover new aspects of viral protein function with importance for pandemic preparedness. This project is part of the HOT-Lab (High Outbreak Threat Preparedness Laboratory) SciLifeLab Technology Development Project, carried out in collaboration with the LU-Fold protein structure prediction infrastructure within the National Bioinformatics Infrastructure Sweden (NBIS), and the Integrated Structural Biology platform (ISB).

The research is closely connected to LU-Fold, a Lund University infrastructure directed by Gemma Atkinson that supports researchers using AlphaFold for high-throughput protein structure prediction. LU-Fold specialises in large-scale prediction of protein complexes and the identification of potential protein–protein interactions.

In short, the group is building infrastructure that could allow researchers to investigate the molecular biology of high-risk viruses before an outbreak occurs and respond more rapidly when new threats emerge.

The Swedish Pathogens Portal is working closely with the Atkinson group to support their efforts. Outputs from their work will be available via the portal in due course.