Epidemic and disease modellers
Write a model as equations; simulate it as individuals, on contact networks and with interventions, and fit it to surveillance data.
SimulatorSystem
State variables
Parameters
Defined quantities
The event log has one row per transition: time, individual, from, to and infector. Together with the scenario and seed it is the full ground truth of the run.
Degree distribution
Contacts of the selected individual
Centre: the selected individual; inner ring: its contacts; outer ring: their contacts. Red lines are transmissions; a dark ring marks isolation. Click a node to move to it.
The contact network (nodes and edges of who can meet whom) is drawn only for a contact structure other than homogeneous mixing: with homogeneous mixing everyone can meet everyone, so the graph shows who infected whom.
The agents are individuals: their epidemic is random, finite and follows the contact structure you chose. The ODE is its mean-field limit (everyone mixes with everyone, infinitely many people). Run many epidemics and see where the agents agree with the ODE, where they do not, and why.
A run is a major outbreak when it infects more than 2% of the population (and more than ten times the initial cases). Medians and 5–95% ranges are over major outbreaks; the shaded band is the 5–95% range of all runs at each time. Seeds 1 to n, so results can be reproduced.
Compare two scenarios of the agent simulation: contact structure, interventions, population and initial conditions, or a changed model. To compare model variants (equations and parameter values) side by side, with sliders and R₀ analysis, use the model lab.
A run counts as a major outbreak when cumulative infections exceed 2% of the initial population. Seeds are 1 to n for both models, so each pair of runs starts from the same random stream.
Elasticity = % change in the outcome per 1% change in the parameter (central difference, ±5%, on the ODE). +1 means proportional. This is a local measure around the current values; switch to Global for LHS-PRCC over parameter ranges.
Detected from the structure of the flows, not from names. Hover a feature for what was found.
Searches 250 million scholarly works with the model's fingerprint as the query. Edit the query to narrow it.
A curated starting list chosen by the model's features, with each title linked to a Scholar search. It is not an exhaustive search of the literature.
Turn the current simulation into realistic surveillance data with a known ground truth: only a fraction of events is reported, each after a random delay, aggregated by day or week, and cases not yet reported by the last simulated day are missing (right truncation).
Run the simulation first (or part of it), then generate.
Columns are matched to the model automatically: new cases are fitted through the observation model, compartment counts (S, E, I, R, deaths, ...) directly. Incidence and the infectious compartments are used by default.
Each simulation runs the full agent model; time grows with the population. Reduce the population for quick exploratory runs.
Counts of the same people on consecutive days are not independent observations; treating them as such (as the Poisson and negative binomial likelihoods do) makes the intervals too narrow. Prefer incidence, or thin the rows.
Equations are exported as written, with the derived individual-level flows listed as comments.
Open platform for epidemic modelling
Kermack is an open-source platform where epidemic and disease modellers, researchers, teachers, students, developers and enthusiasts share and discover epidemic models and datasets, simulate them, and generate synthetic data. The aim is to become the GitHub of epidemic modelling: one place where every model is runnable, comparable and citable.
Works best on a desktop or laptop, where controls and plots sit side by side.
Named after William Ogilvy Kermack, who with Anderson McKendrick wrote the 1927 paper that founded compartmental epidemic modelling.
Write equations, start from the library, or let AI read a paper and write its core model, with every parameter marked as stated in the paper or illustrative.
Duplicate the model in the lab, change an equation or add a parameter, and see every variant on the same axes as you move the sliders.
Run any variant as individuals, with interventions and contact networks, and turn it into surveillance data with under-reporting and delays.
Estimate growth and reproduction numbers, fit the model by likelihood or ABC, and score forecasts on public benchmarks.
Write a model as equations; simulate it as individuals, on contact networks and with interventions, and fit it to surveillance data.
SimulatorTurn a paper's model into a runnable one in a minute, compare variants, and score forecasts on public benchmarks.
Build a lab of model variants and export a handout with plots, questions and an answer key as PDF, Word, LaTeX, Markdown or a Jupyter notebook.
Model labSee what each term of an equation does by moving a slider; every published model in the library comes with its source.
Model libraryAn open-source engine (kermack-core), SBML import, model export to Python and R, and an open API for community models.
ToolboxCurious how epidemics spread and what interventions change? Pick a classic model and watch every individual.
Start exploringEnjoyed using Kermack? Tell us in a sentence or two what it helped you do. With your permission, your words appear here and help other modellers, teachers and students find the tool.
A curated library of published models, from Kermack and McKendrick to COVID-19, malaria, HIV and Ebola, each with its source. Import SBML from BioModels, share any model as a link, and cite it with a stable ID.
Write equations or flows; Kermack derives the individual-level stochastic process whose mean-field limit is your ODE and simulates every individual, under homogeneous mixing or on contact networks, with interventions.
Surveillance data with under-reporting and delays, maximum likelihood and ABC, sensitivity analysis and model comparison, with the ground truth known so methods can be tested.
Upload or paste a case series and get growth rates, doubling times, reproduction numbers, peak timing, epidemic phases and short-term projections, each with its uncertainty, without writing a model. Then see which mechanistic model could explain it.
The Community Modelling and Benchmark Repository: public outbreak and surveillance datasets with documented sources, fixed forecasting and nowcasting protocols, and scores computed the same way for every contribution.
Score-based generative models trained on real surveillance data from three sources produce realistic synthetic epidemic curves, checked against held-out data and for memorisation. Not to be used as real data.
Developer
Mathematician working on stochastic epidemic models, contact tracing on networks and generative AI
Experience, education, awards and the full list of publications are on LinkedIn and Google Scholar.
Kermack is built for the mathematical epidemiology community, and it grows with its users. I welcome collaboration with modellers, epidemiologists, public health teams and students.
Tell colleagues about Kermack
If Kermack contributed to your research, please cite it. If you use a model from the library, please also cite the original paper shown on its card.
Explore and compare
Add or duplicate a model, change an equation or a value, and every variant is redrawn at once. Drag the divider to give the models or the plots more room.
Add a model to start comparing: choose one from the list above and press Add model, bring in the model open in the simulator, extract one from a paper, or write your own.
Reproduction number R(t), all models
Next-generation matrix at the current state (≈ R₀·S(t)/N for simple models). The epidemic grows while R(t) > 1 and declines once it falls below 1.
R₀ as a function of a parameter
Did the lab help your teaching, study or research? ; with your permission it appears on the home page.
Terms and intellectual property
How Kermack treats the papers you read with it, the models it extracts, and the material you export. Short version: Kermack keeps the science (equations, parameter values) with full credit to its authors, never the publication itself, and removes anything an author or rights holder objects to.
Every model says where it comes from: published model (equations from a cited publication), teaching template (written for teaching, not from a paper), lab variant and changed here (modified, so results are not those of the original), and, for extracted models, how many values are stated in the paper and how many were chosen, and whether a person has checked the extraction.
Every model from a paper has a “Report a problem or request removal” button. Authors and rights holders can ask for a model to be corrected or removed; it is hidden while we check, and a removed paper cannot be added again by a new upload.
This page explains how Kermack works; it is not legal advice.
Approve feedback for the home page, and check models extracted from papers. Needs the admin token (the Worker secret ADMIN_TOKEN); it is kept only in this browser tab.
Published and classic infectious disease models, grouped by disease and structure. Every entry states its source and whether its parameter values come from the paper or are illustrative. Open any model in the simulator to run it as an agent-based stochastic process alongside its ODE, add interventions, fit it to data or compare it with another model.
Score-based diffusion models learn what real surveillance curves look like (levels, growth and decline, turning points, short-term variability) from several public data sources. They then generate new curves that follow the same patterns without copying any real place. Compare synthetic and real curves below, and check that the same analysis gives the same answer on both.
You define the dynamics: a compartmental model and its parameters. Kermack simulates every individual with the Gillespie algorithm, so the generating mechanism and the true parameters are known.
A neural network trained on real-world surveillance data generates curves that resemble the data. No mechanism or epidemiological parameters are defined: the curves only reflect patterns in the training data, including its reporting artefacts.
Loading…
Each panel shows a real 64-day curve from a place the model never saw (black) and, in blue, the most similar of the curves just generated. The model does not imitate a particular place: it generates new curves, and this pairing shows that realistic counterparts to real outbreaks appear among them. The memorisation check below confirms they are not copies of training data.
Press Generate.
Synthetic data are not real data. They imitate statistical patterns of the training data, including its biases and reporting artefacts, and describe no real place, period or person. Use them for testing methods, teaching and software development; do not use them as evidence about a real epidemic or as a substitute for surveillance data, and review them carefully before any use. State in any output that the data are synthetic, and cite the model and its training data source.
Median and 10 to 90% range of log(1 + daily cases) at each point of the window.
Distance from each curve to its closest training curve.
Write a model as ordinary differential equations or as flows, and Kermack derives the individual-level stochastic process whose mean-field limit is exactly that ODE. Everything else, from interventions to inference, works on any model you write, paste or import.
Community Modelling and Benchmark Repository
Public outbreak and surveillance datasets, each with its source, its processing and its known problems, and a fixed protocol for forecasting and nowcasting. Run your model here or upload forecasts made elsewhere: every contribution is scored by the platform in the same way and listed with its assumptions, so methods can be compared, checked and improved. This is a shared record of modelling work, not a competition.
Epidemiological insights workspace
Upload, paste or type a case series. Kermack checks the data, then estimates growth and decline, doubling or halving times, reproduction numbers, the peak, epidemic phases, changes in transmission and a short-term projection, each with its uncertainty. No model is needed; when you want one, fit any compartmental model, from a template, the library or your own equations, to see which mechanisms could have produced the curve.
Bars: data. Line: centred moving average. Background: epidemic phase from the growth-rate interval (growth, decline, stable or uncertain, low level). Dashed lines: change points. Right: projection with 50% and 95% intervals.
A compartmental model turns the curve into transmission, recovery and reporting parameters. Several models can fit the same curve equally well, so compare them (AIC) and treat parameters as conditional on the model's assumptions.
Write any system of ODEs (dX/dt = ...) or flows (X -> Y : rate), and describe every state variable and parameter with a comment. Three further kinds run in the model lab: delay models with delay(x, tau) (x at time t − tau), integral (renewal, age-of-infection) models with conv(x, k(s), L) (the integral over s from 0 to L of k(s)·x(t − s)), and discrete-time models written X[t+1] = .... The individual-based simulator runs ODEs and flows; for a kernel there, use a chain of k stages (I1 -> I2 -> ... -> Ik, each at rate k·γ: Erlang-distributed durations).
Data checks, smoothing, peak timing and size. No assumptions about how cases arise.
Local log-linear regressions (quasi-Poisson) for growth rates and doubling times; segmented regression for change points; log-linear projection with Student-t intervals.
Renewal equation: reproduction number by the method of Cori et al. (2013) and its projection; R from the growth rate (Wallinga and Lipsitch 2007). Needs the serial interval.
Any compartmental model (templates, published models or your own equations) fitted by maximum likelihood (negative binomial) with reporting fraction, R0 and intervals; compare models by AIC and open any fit in the simulator.
SBML is the exchange format of the BioModels repository and of COPASI, Tellurium and PySB. Reactions, kinetic laws, local parameters, assignment and rate rules, initial assignments, function definitions and piecewise expressions are converted to equations; the simulator then derives the individual-level flows.
Upload the PDF, the LaTeX source or the text. Kermack reads it and writes the paper's core ODE system in its model language. Every parameter is marked paper when its value is stated in the paper (with where), or illustrative when the paper gives none and a plausible value was chosen.
Published papers whose citation is confirmed by the DOI registry are added to the shared model library, always with that citation and labelled as extracted by AI. Kermack keeps only the model, the parameter values with short notes on where they are stated, and the citation: never the file, its text or its figures. Terms and intellectual property
The paper could not be read right now. Please try again in a few minutes. Meanwhile you can write the model yourself or start from a model in the library.
What did Kermack help you do? A sentence or two is perfect. With your permission, it is shown on the home page after a quick check, with your name and role as you give them.
A request from an author or a rights holder hides the model at once while the moderators check it.
Everything about this run in one document: the model with its source, the scenario, the agents and charts, R₀ and long-term behaviour, and the analyses you have done here (agents against the ODE, comparison of scenarios, parameter sensitivity, surveillance data, fit to data), the infection tree or contact network with related literature. Charts are taken as they are now.
Analyses you have not run yet are left out. The literature search runs automatically if you have not searched.
A handout of this lab for a class, a tutorial or self-study: each model with its source, equations and parameters, the plots as they are now, a summary, questions for discussion and an answer key. Every file carries the Kermack signature, a link that reopens this lab, and the citations.