Malaria parasites are evolving resistance to frontline drugs across Africa, and the computer models that track this spread are becoming too slow and complex for policymakers to use in real time. This project trains deep neural networks—a type of artificial intelligence—to act as fast, accurate stand-ins for those cumbersome simulations. Current disease models require hours or days to run a single scenario, forcing health officials to rely on fixed assumptions rather than exploring the full range of possible futures. The researchers will first test different network architectures against a variety of infectious disease models to establish how much training data is needed for reliable results. They will then apply the best-performing approach to an existing malaria drug resistance model, creating surrogates that can rapidly test thousands of parameter combinations. If successful, national malaria control programmes in Africa will gain a practical tool for scenario planning—quickly comparing the likely trajectories of resistance spread under different intervention strategies, rather than waiting weeks for a single answer. This is applied computational science with a direct policy payoff.
View original technical description
Infectious disease modelling is increasingly complex, integrating novel data and insights into biology and behaviour. Deep neural networks offer a promising method to develop surrogate models for these complex simulations, quickly and at scale. This approach, while established in other sciences, is underutilised in infectious disease research. Our proposal aims to bridge this gap with two objectives: 1. Leverage the advancements in deep learning to provide tools and scientific theory on how to quickly and scalably emulate infectious disease models and deploy them for global health policy. 2. Applying deep learning surrogates to address the emerging threat of malaria drug resistance in Africa. Through these two aims, we will generate both the toolkit and scientific insights necessary for their effective training and subsequent use for global health policy. In the first aim, we will explore different neural network architectures for training surrogate models. Different architectures will be tested against a range of infectious disease models allowing for a functional relationship between infectious disease model complexity and the training data size (infectious disease model simulations) required to train accurate surrogate models. Building on this, the second objective involves developing surrogates for an established malaria drug resistance model. These surrogates will enable exploration of parameter uncertainty in model fits more comprehensively than current methods, which often rely on fixed assumptions. Resultant model fits will provide a more complete understanding of trajectories for the spread of resistance while also feeding into national malaria control program scenario planning exercises.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know