Modeling Malaria Transmission Using Genomic, Immunological, and Epidemiological Data

Despite decades of global effort, malaria remains a persistent public health threat, particularly in endemic regions. One of the primary challenges in controlling and ultimately eliminating malaria is the genetic diversity of the Plasmodium parasite, which enables it to evade host immune responses and reduce the efficacy of interventions such as vaccines. Additionally, the complex interplay between the host, pathogen, and environment necessitates a systems-level approach to understand and predict disease transmission dynamics. This project proposes an integrated modeling framework that combines mechanistic insights from classical mathematical models with the representational power of deep learning to address this challenge.;;The core objective is to build interpretable, predictive models for malaria transmission and control by integrating heterogeneous datasets, including parasite genomic sequences, host immune response profiles, and spatiotemporal epidemiological data. Specifically, the framework will leverage multi-modal deep learning architectures capable of processing high-dimensional inputs from diverse sources. Techniques such as graph neural networks will be employed to model spatial and genetic relationships across populations, while variational autoencoders and contrastive learning will be used for unsupervised feature extraction. To ensure biological plausibility and interpretability, physics-informed neural networks (PINNs) will incorporate mechanistic constraints derived from epidemiological models like SEIR.;;Our aim is to address following questions: how to predict malaria transmission risk using multi-modal data; whether neural networks can capture genotype-phenotype relationships driving immune evasion; and how to interpret complex deep learning models in a biologically meaningful manner. To this end, the project will use explainable AI methods such as SHAP values, integrated gradients, and attention mechanisms. Rigorous model validation will be conducted using spatial and temporal cross-validation and benchmark comparisons with classical models.;;The expected outcomes include a suite of deep learning tools to predict transmission risk, immune escape, and vaccine efficacy, as well as an improved understanding of the genomic and immunological features associated with malaria outcomes. Ultimately, the project aims to support public health decision-making by providing interpretable models and visualizations that inform targeted interventions and resource allocation. This integrative approach holds promise not only for malaria but also as a blueprint for data-driven modeling of other infectious diseases.