Leveraging AI structural inference to inform viral surveillance and vaccine selection

Diseases such as influenza and SARS-CoV-2 constantly evolve to evade population immunity, reducing vaccine effectiveness. At any given time, hundreds of thousands of distinct viral strains circulate, each carrying unique genetic changes. Because of the enormous diversity of viral strains, only a small subset of viruses can be monitored experimentally to guide necessary vaccine updates.

Machine learning models co-developed at the Faculty of Engineering and Information Technology, in collaboration with the WHO Collaborating Center (WHO CC) and Research on Influenza at the Victorian Infectious Diseases Reference Laboratory (VIDRL), have demonstrated the capability of computational models to predict immune escape based on genetic sequence. However, predicting the effect that completely novel genetic changes will have on escaping immunity remains a challenge and key limitation.

This collaboration with MDAP aims to develop next-generation models that reliably detect when unseen genetic changes cause immune escape, using historical and recent influenza surveillance data.

The research team will leverage MDAP's expertise to:

• Integrate AI genetic-structural encodings into current models and provide expertise in state-of-the-art structural encoding methods and guide pipeline implementation
• Pilot development and implementation of alternative AI model architectures that can jointly learn relevant biochemical structural mutation effects, and serum-specific immune target variation
• Design and build pilot a platform presenting model insights for WHO CC partners, translating computational advances into real-world public health impact.

Extending models to predict novel immune escape properties before emergence represents a significant scientific advance. This would enable early detection of variants of concern and proactive surveillance by anticipating viral change rather than following it.

Establishing a pilot platform will bridge methodological development and public health application, accelerating translation of computational advances into tangible improvements in pathogen surveillance, evolution prediction and vaccine strategy, reducing vaccine mismatch and substantially improving vaccine effectiveness and public health outcomes.