ai-powered diagnostic workflow for predicting transmissibility and drug resistance in avian influenza viruses

Emergent diseases such as COVID and avian influenza presented formidable challenges to global public health. To survive and spread, pathogens evolve through genetic mutations which allow them to jump from one host to another or develop resistance to antiviral drugs. Traditional methods for identifying these mutations are slow, costly, relying on detection of actual cases to trigger analysis and response efforts. Alternatively, gain of function studies present biosafety concerns and risks. By contrast, computational methods offer a faster, safer and more cost-effective way to detect mutations underlying phenotypes such as increased transmissibility or antiviral resistance(AVR), providing a proactive approach to combating emerging zoonotic threats. While structure-guided machine learning techniques exist for predicting impacts of mutations, they typically handle single-point mutations, whereas often multiple mutations are required for full phenotypes. Additionally, these existing techniques do not yet employ Protein Language Models (pLMs), which have revolutionised our understanding of protein sequences and structures. These models are trained on vast amounts of protein data and encode evolutionary, structural, and functional information. In this project, we will leverage pLM embeddings, structural information and other features to produce two AI-powered predictors reporting the impacts of mutations likely to induce changes in the affinities of host-viral interactions and AVR. We will use the avian influenza virus as our test target. Avian influenza is carried in wild birds but can be transferred into domestic poultry. Recent outbreaks of H5N1 avian influenza have occurred in 67 countries, leading to the loss of over 131 million chickens and drawing attention to avian-to-mammal and potential mammal-to-mammal transmission. Indeed the first documented cases of cow-to-human transmission in the US in April 2024, highlight the growing risk of zoonotic spillover. Additionally, AVR has already been observed for almost all antivirals against influenza, underscoring the need for proactive intervention and improved understanding of the evolutionary potential of this virus. Our project has six aims to address key challenges in understanding and combating emerging avian influenza viruses, giving computational approaches which can then be extended to other pathogens- Develop a computational pipeline for collating all structural data on pathogen strains and host-pathogen interactions.  Build two AI-based predictors exploiting structural/physicochemical properties and pLM Host-viral interaction predictor reporting mutations that facilitate cell entry, replication etc.  AVR predictor for resistance to the licensed antiviral drugs Validate the AI based predictors using surrogate viral systems Sialic acid binding assay using pseudotyped viruses, expressed HA protein and cell-cell fusion assay Polymerase assays using minigenome reporters performed in human cells (viral polymerase with human ANP32 proteins) AVR to antiviral drug by performing surface plasmon resonance experiment and minigenome polymerase assay Establish a diagnostic portal, reporting impacts and thereby aiding in the identification of novel threats  Organise a Southeast Asia stakeholder workshop to demonstrate the diagnostic portal and engage with experts in zoonotic surveillance and AVR. Once validated, the mutations identified will be seamlessly integrated into global avian influenza surveillance programs, particularly in Southeast Asia, to help combat the spread of resistant infections and reduce infection rates in healthcare settings and broader community. Study results will be useful for development of future antiviral therapeutics with molecules that circumvent resistance development. This project benefits from a collaborative partnership between the UK and Malaysia, promoting knowledge sharing, resource exchange, and adoption of best practices.