What is BioSim Workbench?
BioSim Workbench is a browser-based simulation bench for biochemical reaction networks. You define compartments, species, reactions and rate laws, then simulate the network deterministically with adaptive RK45/BDF integration, stochastically with the exact Gillespie algorithm, fit kinetic parameters to measured data, and quantify how sensitive each output is to every parameter — all on your own machine, with no installation and no server.
- Reaction networks
- Compartments, species, reactions and rate laws: mass action, Michaelis–Menten with competitive inhibition, or a custom expression. Three worked presets are included — enzyme mass action, competitive inhibition, and batch fermentation with Monod growth.
- Deterministic solver
- Adaptive Dormand–Prince RK45 for normal systems, variable-step BDF2/BDF1 for stiff ones, and an LSODA-style auto mode that switches between them when step rejections reveal stiffness. Relative and absolute tolerances are editable.
- Stochastic solver
- Exact Gillespie direct-method simulation with up to 200 replicates, reported as mean ± standard deviation over the ensemble, seeded so a run can be reproduced.
- Steady state
- Damped Newton iteration with a finite-difference Jacobian, started from your initial condition or the end of the last time course, reporting iteration count and final residual.
- Parameter estimation
- Levenberg–Marquardt and Nelder–Mead for local least squares, plus a genetic algorithm and particle swarm for global search, with per-parameter bounds and approximate confidence intervals from the linearised residual covariance.
- Sensitivity and scans
- Forward finite-difference sensitivities ranked by normalised (elasticity) value, and linear or logarithmic parameter scans that plot an output against a swept parameter for dose–response style questions.
- Import and export
- Round-trip a model as JSON, export SBML for interoperability, download time-course and ensemble results as CSV, and generate a PDF analysis report with the model, methods, result tables and figures.
- Privacy
- There is no backend. Model definitions, uploaded CSV data, simulations and reports stay in the browser tab and are never transmitted or stored.
Loading BioSim Workbench…
The numerical engine is about 900 kB and runs entirely on this device.
About the BioSim Workbench
Most systems-biology modelling tools are desktop programs. You install MATLAB or Python bindings, get a licence or a container working, and only then can you ask a question about your network. BioSim Workbench removes that step: it opens in a browser tab and the numerics run locally, which makes it usable on a locked-down bench machine, a borrowed laptop or a tablet in a meeting.
The workflow it supports is the one that actually gets used day to day. Build or import a network, integrate it to get time courses, check whether it settles to a steady state, simulate single-molecule stochastic behaviour when copy numbers are low, fit rate constants to experimental data, and rank which parameters the results depend on. The same model moves between those steps without re-entering anything.
It is deliberately a workbench rather than a pipeline. There is no queue, no batch runner and no server, so nothing about your model, your measured data or your results is transmitted anywhere. That also sets the boundary: it is a working bench for networks you can describe in the editor, not a repository of every published model, and it is not a validated substitute for a clinical or regulatory pipeline.
What this tool does
Reaction-network editor
Define compartments with sizes, species with initial amounts, reactions with equations and rate laws, and reusable parameters. The editor validates the network as you type and flags stoichiometric or kinetic problems before you integrate anything.
Deterministic time courses
Integrate with adaptive RK45, implicit BDF, or the LSODA-style auto mode that detects stiffness from rejected steps and switches solver. Adjustable relative and absolute tolerances and output resolution, with CSV export of the full grid.
Exact stochastic simulation
Run the Gillespie direct method to capture copy-number noise, repeat it as an ensemble, and read the mean ± standard deviation band. Useful when a species sits at tens or hundreds of molecules and the mean-field answer stops being the right one.
Parameter estimation from data
Paste or load experimental CSV data, choose which parameters are free, set bounds, and fit with Levenberg–Marquardt or Nelder–Mead for local least squares, or a genetic algorithm or particle swarm when the landscape is multimodal. Standard errors and confidence intervals come from the residual covariance.
Sensitivity ranking and parameter scans
Rank parameters by normalised sensitivity so effects are comparable across units, then sweep one parameter over a linear or logarithmic range to produce the dose–response curves that tell you whether a fitted value is well constrained or merely unidentifiable.
Interchange and reporting
Import and export models as JSON, export SBML for other tools, download results as CSV, and generate a PDF report containing the model definition, the methods and tolerances actually used, result tables and figures.
Frequently asked questions
Do I need to install Python, R or MATLAB to use this?
No. Everything runs in the browser tab: the ODE solver, the Gillespie simulator, the optimisation algorithms and the sensitivity analysis are all compiled into the page. There is nothing to install, no licence and no runtime to configure.
Are my models and experimental data uploaded anywhere?
No. There is no backend for this tool. Your network definition, any CSV you load, every simulation and the report you generate stay in browser memory and are discarded when you close the tab. That makes it usable for unpublished models and pre-patent work.
When should I use the stochastic tab instead of the deterministic one?
Use the deterministic time course when copy numbers are high, so the concentrations are effectively continuous and the mean-field ODE solution is accurate. Switch to the stochastic tab when a species is present in tens or hundreds of molecules, because genuine reaction-order fluctuations then matter and the mean-field curve hides them. The ensemble view shows how wide that spread is.
What is the difference between RK45 and BDF, and which should I pick?
RK45 is an explicit high-order method: fast and accurate for well-behaved systems, but it needs very small steps to stay stable on stiff ones. BDF is implicit, costs more per step, and handles stiffness that would force RK45 into an unfeasible number of steps. Leave the method on auto unless you have a reason: it runs RK45 and switches to BDF when rejected steps indicate stiffness.
Can I take my model to COPASI, CellNOpt, AMICI or BioModels?
Models export as SBML and as JSON, so they can be moved into tools that read SBML. Two honest limits: this workbench imports its own JSON format rather than parsing arbitrary SBML files, and it does not implement the advanced analysis suites of the desktop packages — metabolic control analysis, time-scale analysis (ILDM, CSP) and Lyapunov exponents are out of scope here, as is automation from a command line.
Is this a replacement for a desktop modelling suite?
For building a network, simulating it, fitting rate constants and checking which parameters matter, it covers the workflow directly and with nothing to install. It is best treated as a fast scratch bench and teaching tool rather than as production infrastructure: keep a local install for large batch studies, ODE code generation and the analysis families listed above.
Related tools
All processing happens locally in your browser. Nothing you upload is transmitted or stored, as described in the privacy policy.