Prepare an SBML model, choose a simulation window and inspect the saved numerical time course. Each method has specific model and input requirements.
1. Load a self-contained model and check route availability
Use the launcher’s Load control for an SBML XML model, then open Genetic Circuit in Synthetic Biology. Preserve the exact model bytes, model identifier, SBML level/version, species identifiers, unit definitions, rules, events and initial conditions. The input must be well-formed SBML; an XML extension alone does not establish that.
The current intake permits up to 10 MiB, 50,000 XML elements, 256 species, 4,096 reactions and 4,096 parameters. It does not resolve external entities and rejects document-type declarations. Composition submodels require an explicitly flattened, self-contained model; this upload path does not silently combine external models. Validation diagnostics may remain as warnings even when the model executes: inspect them.
Summary and Dynamics are available after successful SBML input validation. Other labels remain conditional: Feedback Loop needs a finite full Jacobian, Parameter Sensitivity needs valid settable parameters, and Repressilator requires the annotated BioModels architecture with PX/PY/PZ states. An unavailable method does not receive an invented result. The reviewed three-reference workflow selected Circuit Summary and Circuit Dynamics; it did not validate every Genetic Circuit method.
2. Separate initial time from output time
Model initial time is zero. Initial conditions are applied there. First output time and Last output time select which portion of the resulting trajectory you request. A positive first output time does not move the unevolved initial state to that time.
For example, the reviewed decay model has S(0)=2 and rate dS/dt=−0.5S. Requesting First output time 5, Last output time 10 and Points 51 yields times 5.0, 5.1, …, 10.0. The first value is approximately 0.164170, from 2 exp(−0.5×5), rather than 2. The last is approximately 0.013476. Units in this example are seconds and moles; use the uploaded model’s declared units for other models.
The corrected workflow evolves continuously from zero, preserving pre-window event history and pending delayed assignments, and returns only the requested output rows. Its internal origin row is not an extra customer output point. Parameter perturbations also begin at model time zero. The later revision-8 repair checks the delayed-event cases below; its finite floating-point time rule does not guarantee arbitrary event resolution.
3. Set the output window and numerical controls
| Native control | Contract and interpretation |
|---|---|
| First / Last output time | Finite values, first ≥ 0, last > first. Last output time must not exceed 10⁹ model time units. The work bound applies from time zero, not only to the visible window. |
| Points | An integer from 50 to 10,001, including the two requested endpoints. These are output samples, not the solver’s internal step count. More output points alone do not prove convergence or resolve an event. |
| Rel tol / Abs tol | Relative tolerance from 10⁻¹² to 10⁻²; absolute tolerance from 10⁻¹⁵ to 10⁻³. The three published checks used 10⁻⁹ and 10⁻¹² respectively. Retain settings and compare convergence for the model; these values are not universal accuracy guarantees. |
| Max internal step (0 = auto) | A finite nonnegative value; zero requests automatic steps. For an explicit step cap h, last output time / h must be ≤ 200,000. A narrow late output window still includes its earlier history in this limit. A solver step cap and Points serve different purposes. |
| Integrator | The reviewed checks use CVODE. The native read-only menu displays the configured integrator; it is not a promise that arbitrary solvers can be switched in that control. Unsupported solver requests are refused. Preserve the exact engine and settings with the result. |
| Event order seed | When supported by the checked runtime, use a nonnegative decimal integer up to 9,223,372,036,854,775,807. Retain it as exact text. The seed controls reproducible event-order history, not evidence of a stochastic reaction simulation or biological uncertainty. Unsupported reproducible seeding is refused. |
Select or verify 1–32 output species by identifier. With no explicit selection, the current code uses the first 32 floating species and warns if more exist. Therefore “complete numeric export” means every requested time for the selected outputs, not every species in an arbitrarily large model. Review the selected list before running.
Keep the source and settings unchanged while reviewing one run. The contract refuses unexpected output shape, nonfinite values, a nonmonotonic time axis and materially negative species values. It also refuses a trajectory constant across the entire requested window; that refusal does not prove a biologically invalid model. A partially constant selection produces a warning. Review run status and per-method outcomes rather than treating a failure or unavailable route as zero-valued output.
Current revision 9: boundary species and stoichiometric coefficients
The 2026-11-06 review binds the current app source to engine 2.8.0+quoneon.nleqfree9. Four saved packaged native/local-server examples exercise reactant or product coefficients controlled by rate or assignment rules. Their 51-row exports end at 2.7, 2.3, 1.3 and 1.7 respectively (rounded here), retaining 40 export artifacts. They are analytical examples with explicit models and settings.
The native published-case receipt supports 01736, 01737 and 01738 for the selected floating species S1 and S2 only. Their published grids contain 66 distinct floating-species values, checked on three repeated native runs for 198 comparisons including repeats; three 51-row saved timecourses retain 30 artifacts. Boundary species S3 and S4 are excluded from these app comparisons.
The revision-9 evidence retains 70 export artifacts in total: 40 from the four analytical workflows and 30 from the three supported native published-case workflows. The boundary-only refusal contributes no exports. These two groups keep their separate models, settings and receipts.
Boundary-only case 01636 remains unsupported by this app. It passes the separate engine check but is refused before Run because the app requires floating-species output. The app does not gain boundary-species output selection from the engine repair. The engine review separately covers all published variables for the four repaired cases (242 values); that count must not be presented as native GUI coverage.
The revision-9 review reports 138 affected source tests with no skips or expected failures, a byte-identical 96-object rebuild, and semantic-suite statuses of 1,574 pass, 181 errors, 34 numerical mismatches and 34 unexercised cases across 1,823 cases. The overlapping candidate checks are not additional independent coverage. No previously passing case regressed in the saved comparison. These are recorded review results, not a new suite run performed for this guide; the unresolved statuses and delivery limitations remain.
6. Save the full numeric timecourse
- Inspect the plot’s time range, selected species and units, then open SBML Timecourse. The current output is labeled Species amount; do not silently reinterpret it as concentration or normalize by an unstated compartment volume.
- Export the numeric table and accompanying SBML Timecourse — Method metadata. Preserve the model ID, output units, integrator, model initial time, first/last output times, tolerances, maximum internal step and exact seed.
- Reopen
SBML_Timecourse.csvand verify the requested number of rows and species columns. The 5–10 example has 51 rows, not 52. The full table retains solver output without display rounding or downsampling; the summary’s rounded terminal values are not a substitute. - Keep figures, QC tables, analysis report, run audit and artifact manifest alongside the original SBML and settings. Check file sizes and SHA-256 digests against the manifest. Exact file reopening is not a claim of an editable, reloadable scientific session.
- State the exact accepted engine revision and remaining numerical-time limits with results that depend on delayed events. Record any solver/step/seed changes and their effect rather than silently replacing a prior trajectory.
Read the dated SBML source and evidence inventory. The guide distinguishes the earlier time-origin evidence, revision-8 event evidence and revision-9 boundary evidence, including the app’s floating-species limitation. This guide refresh checks saved receipts and exports; it does not rerun the scientific suite, change an installed runtime, recapture video or establish deployment.