Universal Infectious Disease Forecasting with Mantis 2
Abstract
Public health agencies need to anticipate how outbreaks will unfold, when they will peak, and how long demand on health services will remain elevated. We previously developed a foundation model, Mantis, which was trained entirely on simulated outbreaks---enabling us to generate the diversity and scale of training data needed to support forecasting across diseases. However, our previous work left unresolved when simulation improves predictions and how it should complement observational data. Here, we varied the balance of simulated and observed training data while holding forecasting tasks fixed and varying the available surveillance history. We found that simulation-rich training was particularly valuable with limited local history, and its relative advantage diminished as more observations became available. We also found that simulation-based training reduced exceptionally large forecast errors, suggesting that mechanistic knowledge can help compensate for limited local evidence. Building on these findings, we developed Mantis 2, which combines simulated and observational training to jointly forecast up to six variables sixteen weeks ahead. Across 30 forecasting tasks, Mantis 2 achieved the best aggregate probabilistic accuracy over short, medium, and long horizons among the machine learning, statistical, and baseline models evaluated. It missed roughly 30% fewer outbreak peaks than the strongest comparison model, with successfully anticipated peaks identified more than six weeks in advance on average. We publicly release Mantis 2, enabling public health agencies and researchers to forecast disease outcomes without developing a separate model for each disease or location.
Figures & tables
| In-domain evaluation | |||
|---|---|---|---|
| Disease | Location | Transmission | Signals |
| Acute diarrhea [ 26 ] | Thailand (Provincial) | Enteric | Univariate |
| Enteric fever [ 26 ] | Thailand (Regional) | Enteric | Univariate |
| Hepatitis A [ 26 ] | Thailand (Regional) | Enteric | Univariate |
| Typhoid fever [ 27 ] | USA (Regional) | Enteric | Univariate |
| Salmonellosis [ 27 ] | USA (Regional) | Enteric | Univariate |
Appendix figures & tables8 assets
Supplementary material from the paper’s appendix.
Appendix
| Quantity | Prior | Notes / clipping | Citation |
|---|---|---|---|
| Infectious period (d) | clip | [ 36 , 37 ] | |
| Latent period (d) | clip | [ 38 , 39 ] | |
| Immunity duration (d) | clip | [ 40 , 41 ] | |
| (asymptomatic) | mean | [ 42 , 43 ] | |
| (asymp. trans.) | mean | [ 44 ] | |
| ( ) |
| Mechanism | Quantity | Prior | Cite |
| all | (base) | (mean ) | [ 46 , 47 ] |
| (vector) | (mean ) | [ 48 ] | |
| (environmental) | (mean ) | [ 49 ] | |
| vector | [ 48 ] | ||
| [ 48 ] | |||
| (biting) | [ 48 ] |
| Model | Weeks 1–4 | Weeks 1–8 | Weeks 1–16 |
|---|---|---|---|
| Mantis 2 | 0.20913 | 0.28439 | 0.36965 |
| Chronos-2 | 0.23097 | 0.31084 | 0.39762 |
| Persistence | 0.28046 | 0.36928 | 0.46776 |
| LSTM | 0.41975 | 0.45859 | 0.51038 |
| Seasonal naive | 0.52003 | 0.52281 | 0.52770 |
| SARIMA | 0.59501 | 0.62158 | 0.65710 |
| Horizon | Model | 50% coverage | 95% coverage | Datasets | |
|---|---|---|---|---|---|
| Weeks 1–4 | Mantis 2 | 48.59 | 96.14 | 164,095 | 30 |
| Chronos-2 | 49.14 | 96.16 | 164,095 | 30 | |
| Persistence | 50.72 | 90.77 | 164,095 | 30 | |
| LSTM | 40.59 | 85.54 | 164,095 | 30 | |
| Seasonal naive | 52.63 | 90.73 | 133,006 | 29 | |
| SARIMA | 64.60 | 93.94 | 158,219 | 30 |
| Dataset | Mantis 2 | Chronos-2 | Persistence | LSTM | Seasonal | SARIMA | ETS |
|---|---|---|---|---|---|---|---|
| COVID-19 | 0.70492 | 0.70515 | 0.65241 | 1.11491 | 0.76370 | 0.73894 | 0.96699 |
| Dengue–Ecuador | 0.28054 | 0.28172 | 0.39999 | 0.54665 | 0.51285 | 0.40595 | 0.41814 |
| Dengue–Indonesia | 0.21717 | 0.22298 | 0.22676 | 0.46587 | 0.45524 | 0.18264 | 0.20858 |
| Dengue–Nicaragua | 0.41206 | 0.42872 | 0.40332 | 0.56666 | 0.65005 | 0.41656 | 0.83064 |
| Dengue–Singapore | 0.30825 | 0.31734 | 0.29664 | 0.56918 | 0.81328 | 0.31197 | 0.68681 |
| Dengue–Thailand | 0.44336 | 0.52093 | 0.58610 | 0.61893 | 0.87349 | 0.57618 | 0.80000 |
| Dataset | Mantis | Mantis 2 | Improvement (%) |
|---|---|---|---|
| Dengue—Ecuador | 0.24591 | 0.19890 | 19.12 |
| Dengue—Indonesia | 0.17306 | 0.10400 | 39.91 |
| Dengue—Nicaragua | 0.34377 | 0.30749 | 10.55 |
| Dengue—Singapore | 0.24938 | 0.20207 | 18.97 |
| Dengue—Thailand | 0.42178 | 0.28882 | 31.52 |
| Candida auris —U.S. | 0.35623 | 0.31928 | 10.37 |
| Horizon range | Base WIS | Adjusted WIS | Improvement (%) |
|---|---|---|---|
| Weeks 1–4 | 0.213999 | 0.216261 | -1.06 |
| Weeks 1–8 | 0.295455 | 0.298283 | -0.96 |
| Weeks 1–16 | 0.392063 | 0.394660 | -0.66 |
| Ensemble | Weeks 1–4 | Weeks 1–8 | Weeks 1–16 |
|---|---|---|---|
| All nonempty subsets | -0.03% | -0.07% | 0.18% |
| Subsets with at least two variables | 0.15% | 0.14% | 0.27% |
Epidemic Forecasting
Epidemic forecasting predicts disease incidence, outbreaks, and healthcare burden across time and regions to guide surveillance and public-health decisions.
See all 40 papers