SERVICES / INDEPENDENT COMPUTATIONAL ANALYSIS
Make the assumptions
visible.
Mathematical modeling and simulation for biotech, pharmaceutical and research teams. Each engagement starts with one question and a paid, defined scope.
01 / SERVICE
Global sensitivity analysis
Which inputs actually drive the result?
A model may depend on many parameters, while only a few materially change the outcome you care about. Testing one parameter at a time can miss interactions and changes across the plausible input space.
The analysis
Define the outcome and defensible parameter ranges, then screen or quantify influence across those ranges. Depending on the model and compute budget, methods may include Morris screening or variance-based Sobol indices. Correlated inputs require an appropriate sampling and analysis plan.
What you receive
A ranked set of influential inputs, interaction findings where supported, convergence checks, and figures showing where better measurements could be useful.
What is needed to begin
An existing executable model, a defined outcome, and justified ranges or distributions for uncertain inputs.
02 / SERVICE
Parameter identifiability
Can the observations distinguish the mechanisms?
Different parameter combinations can produce nearly the same fitted curve. A close fit alone does not establish that each parameter, or the mechanism it represents, has been resolved.
The analysis
Assess structural identifiability where feasible and practical identifiability under the actual sampling and noise assumptions. Use tools such as profile likelihood, simulation-based recovery and parameter-dependence analysis as appropriate.
What you receive
A map of parameters or combinations the data can constrain, competing explanations that remain plausible, and candidate measurements to separate them.
What is needed to begin
Model equations or code, observation definitions, sampling times and suitable data or a defensible simulated-data scenario.
03 / SERVICE
Uncertainty propagation
How much confidence belongs around a prediction?
A single fitted trajectory can hide the consequences of uncertain parameters, variable measurements and alternative model assumptions.
The analysis
Carry justified input uncertainty through the model using Monte Carlo or other suitable sampling methods. Separate parameter uncertainty from measurement noise and compare alternative model structures when the scope supports it.
What you receive
Prediction ranges with their assumptions stated, scenario comparisons, and an explanation of which uncertainties materially change the interpretation.
What is needed to begin
A model, a prediction of interest, and evidence for parameter distributions or a clearly labeled set of plausible scenarios.
04 / SERVICE
Model development & simulation
What could happen under different conditions?
A scientific hypothesis is easier to examine when its assumptions are explicit and its predicted behavior can be reproduced.
The analysis
Develop or adapt a bounded mathematical model, implement it in reproducible code, check numerical behavior, and compare defined scenarios. Calibration and checks against held-out observations are included when suitable data are available and agreed in scope.
What you receive
Documented model equations, runnable simulations, numerical checks, scenario figures and a clear account of what the model can and cannot support.
What is needed to begin
A specific biological question, relevant mechanism or literature, and data availability sufficient for the intended use.
05 / SERVICE
Simulation & experimental design
Which measurement would reduce useful uncertainty?
Additional data are most useful when they distinguish explanations or constrain the prediction that matters. More measurements at the same time points may leave the main uncertainty unresolved.
The analysis
Simulate candidate sampling schedules, readouts or experimental conditions under plausible scenarios. Compare expected parameter precision, prediction uncertainty or ability to distinguish mechanisms, subject to practical constraints.
What you receive
A comparison of feasible designs, the tradeoffs behind their ranking, and a proposed next measurement with its assumptions and limitations.
What is needed to begin
A suitable model, candidate measurements, noise assumptions and practical limits such as sample count, timing and assay availability.
Start with one
bounded analysis.
A first project can be an independent review of one model, one sensitivity study, or one comparison of sampling schedules. Deliverables include reproducible code, clear figures and a short report covering results and limitations.
Scope, fee, timing and data access are agreed before work begins. Method selection depends on feasibility; every project does not need every method.
Discuss a paid project ↗