Advancing Real-time Pandemic Forecasting Using Large Language Models: A COVID-19 Case Study (preprint)

Forecasting the short-term spread of an ongoing disease outbreak is a formidable challenge due to the complexity of contributing factors, some of which can be characterized through interlinked, multi-modality variables such as epidemiological time series data, viral biology, population demographics, and the intersection of public policy and human behavior. Existing forecasting model frameworks struggle with the multifaceted nature of relevant data and robust results translation, which hinders their performances and the provision of actionable insights for public health decision-makers. Our work introduces PandemicLLM, a novel framework with multi-modal Large Language Models (LLMs) that reformulates real-time forecasting of disease spread as a text reasoning problem, with the ability to incorporate real-time, complex, non-numerical information that previously unattainable in traditional forecasting models. This approach, through a unique AI-human cooperative prompt design and time series representation learning, encodes multi-modal data for LLMs. The model is applied to the COVID-19 pandemic, and trained to utilize textual public health policies, genomic surveillance, spatial, and epidemiological time series data, and is subsequently tested across all 50 states of the U.S. Empirically, PandemicLLM is shown to be a high-performing pandemic forecasting framework that effectively captures the impact of emerging variants and can provide timely and accurate predictions. The proposed PandemicLLM opens avenues for incorporating various pandemic-related data in heterogeneous formats and exhibits performance benefits over existing models. This study illuminates the potential of adapting LLMs and representation learning to enhance pandemic forecasting, illustrating how AI innovations can strengthen pandemic responses and crisis management in the future.

Combining genomic and epidemiological data to compare the transmissibility of SARS-CoV-2 lineages (preprint)

Emerging SARS-CoV-2 variants have shaped the second year of the COVID-19 pandemic and the public health discourse around effective control measures. Evaluating the public health threat posed by a new variant is essential for appropriately adapting response efforts when community transmission is detected. However, this assessment requires that a true comparison can be made between the new variant and its predecessors because factors other than the virus genotype may influence spread and transmission. In this study, we develop a framework that integrates genomic surveillance data to estimate the relative effective reproduction number (Rt) of co-circulating lineages. We use Connecticut, a state in the northeastern United States in which the SARS-CoV-2 variants B.1.1.7 and B.1.526 co-circulated in early 2021, as a case study for implementing this framework. We find that the Rt of B.1.1.7 was 6-10% larger than that of B.1.526 in Connecticut in the midst of a COVID-19 vaccination campaign. To assess the generalizability of this framework, we apply it to genomic surveillance data from New York City and observe the same trend. Finally, we use discrete phylogeography to demonstrate that while both variants were introduced into Connecticut at comparable frequencies, clades that resulted from introductions of B.1.1.7 were larger than those resulting from B.1.526 introductions. Our framework, which uses open-source methods requiring minimal computational resources, may be used to monitor near real-time variant dynamics in a myriad of settings.