A doctor takes a swab from a woman to test for the COVID-19 virus at a fever clinic in Yinan county in eastern China's Shandong province on Wednesday, Feb. 12, 2020. AP
A study published with the involvement of Oxford Martin School researchers outlines for the first time how advances in AI can accelerate breakthroughs in infectious disease research and outbreak response.
The Nature study – published following last week’s AI Action Summit and amidst increasing global debate on AI investment and regulation – puts particular emphasis on safety, accountability, and ethics in the deployment and use of AI in infectious disease research.
Calling for a collaborative and transparent environment – both in terms of datasets and AI models – the study is a partnership between scientists from the University of Oxford and colleagues from academia, industry, and policy organisations across Africa, America, Asia, Australia, and Europe.
So far, medical applications of AI have predominantly focused on individual patient care, enhancing, for example, clinical diagnostics, precision medicine, or supporting clinical treatment decisions. This review instead considers the use of AI in population health. The study finds that recent advances in AI methodologies are performing increasingly well even with limited data – a major bottleneck to date. Better performance on noisy and limited data is opening new areas for AI tools to improve health across both high-income and low-income countries.
Lead author Professor Moritz Kraemer, from the University of Oxford’s Pandemic Sciences Institute and Oxford Martin Programme on Pandemic Genomics, stated, “In the next five years, AI has the potential to transform pandemic preparedness. It will help us better anticipate where outbreaks will start and predict their trajectory, using terabytes of routinely collected climatic and socio-economic data. It might also help predict the impact of disease outbreaks on individual patients by studying the interactions between the immune system and emerging pathogens. Taken together, and if integrated into countries’ pandemic response systems, these advances will have the potential to save lives and ensure that the world is better prepared for future pandemic threats.”
Opportunities for AI and pandemic preparedness identified in the research include promising advances in improving current models of disease spread, aiming to make modelling more robust, accurate, and realistic. Progress in pinpointing areas of high-transmission potential will help ensure that limited healthcare resources can be allocated in the most efficient possible way.
There is also potential to improve genetic data in disease surveillance, ultimately accelerating vaccine development and the identification of new variants. AI could help determine the properties of new pathogens, predict their traits, and identify whether cross-species jumps are likely. Furthermore, it may assist in predicting which new variants of already-circulating pathogens – such as SARS-CoV-2 and influenza viruses – might arise, and which treatments and vaccines are best in reducing their impact.
Moreover, AI can facilitate the integration of population-level data with data from individual-level sources, including wearable technologies such as heart rate and step counts, to better detect and monitor outbreaks. It can create a new interface between highly technical science and healthcare professionals with limited training, improving capacity in settings that need these tools the most.
However, not all areas of pandemic preparedness and response will be equally impacted by advances in AI. For example, whereas protein language models hold great promise for speeding up understanding of how virus mutations can impact disease spread and severity, advances in foundational models might only provide modest improvements over existing approaches to modelling the speed at which a pathogen is spreading.
The scientists urge caution in suggesting that AI alone will solve infectious disease challenges, but they believe that integrating human feedback into AI modelling workflows might help overcome existing limitations. The authors are particularly concerned with the quality and representativeness of training data, the limited accessibility of AI models to the wider community, and potential risks associated with the deployment of black-box models for decision-making.
Study author Professor Eric Topol, MD, founder and director of the Scripps Research Translational Institute, remarked, “While AI has remarkable transformative potential for pandemic mitigation, it is dependent upon extensive worldwide collaboration and comprehensive, continuous surveillance data inputs.”
Study lead author Samir Bhatt from the University of Copenhagen and Imperial College London added, “Infectious disease outbreaks remain a constant threat, but AI offers policymakers a powerful new set of tools to guide informed decisions on when and how to intervene.” The authors suggest rigorous benchmarks to evaluate AI models, advocating for strong collaborations between government, society, industry, and academia for sustainable and practical development of models for improving human health.