Mantis: A Foundation Model for Mechanistic Disease Forecasting
Summary: arXiv:2508.12260v5 Announce Type: replace
Abstract
Infectious disease forecasting in novel outbreaks or low-resource settings is hampered by the need for large disease and covariate data sets, bespoke training, and expert tuning, all of which can hinder rapid generation of forecasts for new settings. To help address these challenges, we developed Mantis, a foundation model trained entirely on mechanistic simulations, which enables out-of-the-box forecasting across diseases, regions, and outcomes, even in settings with limited historical data.
Key Features of Mantis
Mantis offers several unique advantages when it comes to disease forecasting:
- Data Efficiency: Trained on mechanistic simulations, Mantis does not rely on extensive real-world datasets, making it suitable for low-resource environments.
- Versatile Forecasting: Capable of generating forecasts for various diseases and outcomes, Mantis is designed for adaptability across diverse epidemiological contexts.
- Robust Performance: In comparative evaluations, Mantis outperformed traditional forecasting models, demonstrating its reliability in predicting disease outbreaks.
Performance Evaluation
To assess Mantis, we evaluated it against 78 forecasting models across sixteen diseases with diverse transmission modes. The evaluation focused on two main performance metrics:
- Point Forecast Accuracy: Measured by the mean absolute error, Mantis consistently achieved lower errors than all models in the CDC’s COVID-19 Forecast Hub during backtesting on early pandemic forecasts.
- Probabilistic Performance: Assessed using weighted interval score and coverage, Mantis ranked among the top two models across various diseases.
Generalization Capabilities
One of the remarkable aspects of Mantis is its ability to generalize to diseases with transmission mechanisms not represented in its training data. This capacity indicates that Mantis captures fundamental contagion dynamics rather than merely memorizing disease-specific patterns. Such generalization is crucial in the face of emerging infectious diseases.
Conclusion
The development of Mantis illustrates the potential of simulation-based foundation models in the field of disease forecasting. By providing a general-purpose, accurate, and deployable forecasting tool, Mantis addresses many limitations faced by traditional models, especially in challenging settings. As public health continues to confront novel outbreaks, tools like Mantis will be invaluable in enhancing our forecasting capabilities and improving response strategies.