1. Why Communicating QSP Results Is a Modeling Skill
A quantitative systems pharmacology (QSP) model can contain hundreds of variables, parameters, equations, biological mechanisms, and simulated scenarios. Yet the purpose of presenting the model is rarely to show all of that information at once. The purpose is to communicate what the model contributes to a scientific question.
Good QSP communication therefore requires more than producing attractive plots. The analyst must explain what was simulated, what the model predicts, how certain those predictions are, which assumptions matter, and what conclusions are justified by the analysis.
A useful QSP presentation moves from model structure to simulation results and finally to a clearly stated scientific interpretation.
2. Start With the Scientific Audience
The same QSP analysis may need to be communicated to modelers, pharmacologists, clinical scientists, statisticians, regulatory scientists, or project teams. Each audience needs a different level of technical detail.
| Audience | Usually needs to understand | Technical detail to emphasize |
|---|---|---|
| QSP modelers | Mechanistic assumptions, equations, parameterization, identifiability, and diagnostics | Model structure, parameter uncertainty, sensitivity, calibration, and validation |
| Clinical scientists | What the model predicts for disease, treatment, and patient-relevant outcomes | Biological interpretation, treatment scenarios, response distributions, and limitations |
| Project teams | How the model changes understanding of a development question | Scenario comparisons, key drivers, uncertainty, and implications |
| Regulatory audiences | Whether the evidence is scientifically justified and appropriately qualified | Model purpose, assumptions, verification, validation, uncertainty, and applicability |
A common communication failure is to begin with the model rather than the question. A presentation can spend ten slides explaining compartments and equations before the audience learns what decision the simulation was intended to inform.
3. State the Question Before Showing the Result
A QSP result is meaningful only in relation to a question. Before presenting a plot, state what the simulation is intended to investigate.
Examples include:
- How does target inhibition change downstream biomarker dynamics?
- What degree of target engagement is required to produce a specified biological effect?
- Which mechanisms could explain a difference in treatment response?
- How might changes in dose or schedule affect the predicted response?
- Which biological parameters contribute most strongly to uncertainty in the outcome?
- What experimental observations would most reduce uncertainty in the model?
These questions lead to different analyses. A dose-response question may emphasize scenario simulations, whereas an uncertainty question may emphasize parameter distributions and sensitivity analysis.
This sequence provides a simple organizing principle for both figures and presentations.
4. Clearly Separate Observed Data From Model Predictions
One of the most important principles in communicating mechanistic model results is to distinguish what was actually observed from what was generated by the model.
| Information type | Example | How to describe it |
|---|---|---|
| Observed | Measured biomarker concentration | "Observed concentration" |
| Model-predicted | Simulated biomarker trajectory | "Model-predicted trajectory" |
| Calibrated | Parameter estimated from experimental data | "Estimated/calibrated parameter" |
| Assumed | Fixed biological parameter based on literature | "Literature-based fixed value" |
| Scenario-derived | Predicted response under an untested dose | "Simulated scenario" |
The distinction becomes particularly important when a model is used to extrapolate beyond the experimental conditions used for calibration. A simulated result can be scientifically useful without being an observation.
5. Choose Outputs That Answer the Question
A complex QSP model may generate hundreds or thousands of state variables. Most of these should not appear in the primary communication of results.
The first task is therefore to identify the decision-relevant output. This may be a biomarker, target engagement measure, disease burden, response probability, exposure metric, or other endpoint.
| Question | Potential output | Useful visualization |
|---|---|---|
| What happens over time? | Biomarker or disease-state trajectory | Time-course plot |
| How does response depend on dose? | Endpoint versus dose | Dose-response curve |
| Which mechanism drives the result? | Output sensitivity to parameters | Tornado plot or sensitivity plot |
| How uncertain is the prediction? | Prediction distribution | Prediction interval or uncertainty band |
| How do treatments compare? | Difference or ratio between scenarios | Scenario comparison plot |
| Which biological pathway changes? | Multiple mechanistic outputs | Pathway or mechanism diagram |
A useful output is not necessarily the variable that is most interesting from a modeling perspective. It is the variable that best connects the model to the scientific question.
6. Communicating Dynamic QSP Predictions
Many QSP models are explicitly dynamic. They describe how concentrations, signaling states, cell populations, disease variables, and other quantities change over time. Time-course plots are therefore among the most common QSP visualizations.
A time-course figure should make the distinction between observations and model-generated trajectories immediately apparent.
Time-course plots should identify the units of both axes, the treatment or scenario, and whether lines represent individual simulations, median predictions, means, or other summaries.
If uncertainty is available, it is often more informative to show an uncertainty band than to display a single deterministic trajectory.
7. Communicating Model Uncertainty
QSP predictions are rarely known exactly. Uncertainty can arise from parameter estimation, biological variability, measurement error, structural assumptions, or uncertainty about mechanisms.
For a predicted quantity \(Y\), uncertainty can be represented conceptually as a distribution rather than a single number:
The exact interpretation of a prediction interval depends on how the simulation was constructed. What matters for communication is that the audience can distinguish the central prediction from the range of plausible outcomes represented by the analysis.
| Presentation | What it communicates | Potential limitation |
|---|---|---|
| Single trajectory | Central model prediction | Can hide substantial uncertainty |
| Confidence or credible interval | Parameter or inference uncertainty under a specified framework | Interpretation depends on the statistical framework |
| Prediction interval | Range of future or simulated outcomes | Must be defined clearly |
| Simulation envelope | Range of model-generated trajectories | Can be misinterpreted if simulation assumptions are unclear |
| Individual trajectories | Between-subject or scenario variability | Can become visually crowded |
8. Communicating Sensitivity Analysis
Sensitivity analysis asks how strongly a model output changes when model inputs or parameters change. It can help identify which biological assumptions matter most for a particular prediction.
For an output \(Y\) and parameter \(\theta_i\), a local sensitivity can be represented conceptually as:
Global sensitivity analyses may instead explore parameter distributions over a broad range of plausible values.
A sensitivity plot is most informative when the output, parameter ranges, and direction of influence are explicitly defined.
A sensitivity ranking should not automatically be interpreted as a ranking of biological importance. A parameter may appear influential because its plausible range is wide, because of model parameterization, or because of correlations with other parameters.
9. Communicating Scenario Simulations
Scenario simulation is one of the major strengths of QSP modeling. Once the model has been calibrated and evaluated, investigators can explore conditions that may not yet have been experimentally tested.
Examples include:
- Different dose levels.
- Different dosing intervals.
- Different degrees of target engagement.
- Changes in biological pathway activity.
- Alternative mechanisms of action.
- Different patient or disease characteristics.
- Combination-treatment scenarios.
A scenario comparison should clearly identify the baseline or reference scenario. Without a reference, it can be difficult to understand whether an apparent difference is scientifically meaningful.
For some outcomes, a relative comparison may be more useful:
The choice between absolute difference, relative difference, ratio, or another measure should follow the scientific question rather than the convenience of the visualization.
10. Show the Mechanism Behind the Prediction
QSP models are valuable partly because they provide mechanistic explanations for predictions. A strong presentation therefore does not stop at saying that one scenario produced a larger response.
Instead, explain the causal or mechanistic sequence represented by the model.
Mechanistic communication connects the intervention to intermediate biological processes and ultimately to the endpoint of interest.
This is particularly useful when multiple pathways contribute to an outcome. Showing intermediate variables can make the model's mechanistic interpretation more transparent.
11. Communicating Model Evaluation and Validation
Before using a QSP model to make predictions, audiences need to understand whether the model has been evaluated against relevant evidence.
Model evaluation can include comparison with experimental observations, historical data, external datasets, known biological behavior, or other independent evidence.
| Question | What to communicate |
|---|---|
| Does the model reproduce calibration data? | Show observed data and model predictions under calibration conditions. |
| Does it reproduce independent observations? | Show external or validation data separately from calibration data. |
| Does it reproduce qualitative behavior? | Describe the biological pattern and whether the model reproduces it. |
| Are parameters biologically plausible? | Provide ranges, sources, and rationale where appropriate. |
| Are predictions robust? | Show uncertainty and sensitivity analyses. |
A visually impressive fit is not, by itself, evidence that the model is suitable for every future prediction. The relevance of the validation evidence depends on the intended use of the model.
12. Communicate Limitations Explicitly
A QSP model is a structured representation of biological knowledge. It is therefore useful to communicate not only what the model predicts but also what assumptions constrain those predictions.
Important limitations may include:
- Limited or sparse experimental data.
- Uncertain biological parameters.
- Alternative plausible mechanisms.
- Structural assumptions in the model.
- Uncertainty in disease biology.
- Limited representation of patient heterogeneity.
- Extrapolation beyond the conditions used for model evaluation.
- Correlation or non-identifiability among parameters.
- Uncertainty in the relationship between a model endpoint and a clinical outcome.
A limitation should be described specifically enough that the audience understands how it affects interpretation.
For example, saying "the model has limitations" is much less informative than saying that "the predicted response at high target engagement is sensitive to an uncertain parameter describing pathway turnover."
13. Distinguish Different Sources of Uncertainty
Not all uncertainty in a QSP analysis has the same meaning. Separating sources of uncertainty can substantially improve scientific communication.
| Source | Description | Typical communication approach |
|---|---|---|
| Parameter uncertainty | Uncertainty about numerical parameter values | Parameter distributions, confidence or credible intervals, sensitivity analysis |
| Biological variability | Real differences among individuals or systems | Population simulations or distributions of predicted outcomes |
| Measurement uncertainty | Imprecision in experimental observations | Observation-error representation and data uncertainty |
| Structural uncertainty | Uncertainty about model equations or mechanisms | Alternative model structures or scenario analyses |
| Extrapolation uncertainty | Uncertainty introduced when predictions extend beyond supporting data | Applicability analysis and explicit qualification of predictions |
14. Designing Effective QSP Figures
Figures often carry most of the information in a QSP presentation. A good figure should allow the audience to identify the scientific message without reconstructing the analysis themselves.
Useful design principles include:
- Give every axis a meaningful label and unit.
- Identify treatment, dose, and scenario definitions.
- Distinguish observations from predictions.
- Show uncertainty when it materially affects interpretation.
- Avoid displaying every simulated variable in a single figure.
- Use consistent definitions across related figures.
- State what the shaded region, line, point, or interval represents.
- Include a reference scenario when making comparisons.
The title can also communicate the scientific question. For example, "Predicted biomarker suppression versus dose" is generally more informative than simply "Simulation Results."
15. When a Table Is Better Than a Plot
Not every QSP result needs to be visualized as a curve. Tables can be particularly useful for summarizing scenario definitions, parameter assumptions, or key numerical predictions.
| Scenario | Dose | Target engagement | Predicted response |
|---|---|---|---|
| Reference | Low | 35% | 0.82 |
| Scenario A | Medium | 68% | 0.61 |
| Scenario B | High | 88% | 0.48 |
A table is especially useful when exact numerical values matter or when the audience needs to compare several scenarios simultaneously.
16. Worked Example: Communicating a QSP Dose-Response Simulation
Consider a hypothetical QSP model describing a drug that inhibits a biological target and produces a downstream disease-related response. Suppose the model is used to simulate three dose levels.
| Scenario | Predicted target engagement | Predicted response |
|---|---|---|
| Placebo/reference | 0% | 1.00 |
| Low dose | 45% | 0.82 |
| Medium dose | 70% | 0.63 |
| High dose | 88% | 0.51 |
Step 1: Define the comparison
The reference scenario is the untreated condition with an outcome of 1.00. The predicted effect of treatment can therefore be summarized relative to this reference.
Step 2: Calculate the relative change
For the medium-dose scenario:
The model therefore predicts a 37% reduction in the outcome relative to the reference scenario.
Step 3: Connect the prediction to mechanism
The model predicts that the medium dose produces approximately 70% target engagement. The reduction in the downstream outcome is therefore interpreted through the modeled relationship between target engagement and the disease-related pathway.
Step 4: Add uncertainty
Suppose the simulation produces a 90% prediction interval of 0.52 to 0.75 for the medium-dose response. The result should then be communicated as a central prediction together with the range represented by the simulation, rather than simply as 0.63.
Step 5: State the scientific interpretation
Notice what this statement does not say. It does not claim that the simulation proves the treatment will produce the predicted clinical outcome. Instead, it identifies the model-based result, the mechanism represented by the model, and the conditions under which the prediction should be interpreted.
17. Communicating Differences Between Scenarios
When comparing scenarios, the most useful result is often the difference between them rather than the absolute prediction from each scenario.
For two scenarios \(A\) and \(B\), the absolute contrast is:
A relative contrast can be expressed as:
The appropriate comparison depends on the endpoint. For a bounded biomarker, percentage change may be intuitive. For an event probability or survival-related quantity, another scale may be more appropriate.
Most importantly, define the direction of the comparison. A statement such as "Scenario A is 20% lower" is incomplete unless the reference quantity is clear.
18. Communicating Heterogeneity and Population Simulations
QSP models can be used to explore how differences among simulated individuals affect predicted outcomes. This is particularly useful when biological variability is scientifically important.
Instead of presenting a single deterministic trajectory, a population simulation may produce a distribution:
The resulting distribution can show a median response, percentile range, probability of exceeding a threshold, or other clinically meaningful summary.
However, simulated heterogeneity should not automatically be interpreted as observed clinical variability. The distribution is conditional on the assumptions used to generate the simulated population.
19. Communicating Threshold-Based Predictions
QSP analyses often investigate whether a biological or clinical threshold is reached. Examples include target engagement above a specified level or disease burden below a specified value.
If \(Y\) is an output and \(T\) is a threshold, the quantity of interest may be:
For a simulated population, this can be estimated as:
When communicating such a result, report the threshold, the population or simulation conditions, and the definition of the outcome.
20. Communicating a Result When the Model Predicts Little or No Effect
A QSP model may predict that changing a parameter, dose, mechanism, or intervention produces little change in the endpoint of interest. This can be scientifically useful.
The interpretation should distinguish between several possibilities:
- The mechanism genuinely has little influence on the modeled endpoint.
- The selected parameter range does not span a biologically important regime.
- Another pathway compensates for the perturbation.
- The endpoint is insensitive to the mechanism under the simulated conditions.
- The result is obscured by parameter or structural uncertainty.
A flat response should therefore prompt mechanistic investigation rather than being automatically described as proof that a biological mechanism is unimportant.
21. Communicating Extrapolation Beyond the Data
One of the defining uses of QSP is to explore conditions that have not yet been experimentally observed. Such predictions can be useful precisely because they extend beyond existing data, but that extension should be explicit.
A QSP prediction can extend beyond observed data, but the distinction between data-supported behavior and extrapolation should be visible in the communication.
Extrapolation is not inherently inappropriate. It is one of the reasons mechanistic models are useful. The key is to communicate which portions of the prediction are directly supported by data and which depend more strongly on model assumptions.
22. A Practical Structure for a QSP Results Presentation
A clear QSP results presentation can often follow a simple progression.
- Scientific question. State the decision or biological question.
- Model purpose. Explain what the model is intended to represent.
- Relevant evidence. Summarize the data and knowledge supporting the model.
- Model evaluation. Show how the model reproduces relevant observations.
- Primary prediction. Show the main result directly.
- Mechanistic explanation. Explain why the model predicts the result.
- Uncertainty. Show how much the prediction varies.
- Sensitivity. Identify assumptions or parameters that drive the prediction.
- Scenario comparison. Explain how alternative conditions change the result.
- Limitations. State what the analysis does not establish.
- Conclusion. Give the model-based scientific interpretation in one or two sentences.
23. Use Precise Model-Based Language
The language used to describe QSP results should reflect the evidentiary status of the result.
| Less precise | More precise |
|---|---|
| "The drug reduces disease." | "The model predicts a reduction in the disease-related endpoint under the simulated treatment conditions." |
| "The target causes the response." | "The model attributes the predicted response change to the modeled target-mediated pathway." |
| "The treatment will work." | "The simulation predicts the specified outcome under the model assumptions." |
| "Parameter X is the most important." | "The selected sensitivity analysis identifies parameter X as highly influential for this output." |
| "The model proves the mechanism." | "The model provides a mechanistic explanation consistent with the available evidence." |
This distinction is not merely linguistic. Precise wording prevents model-generated evidence from being mistaken for direct experimental or clinical evidence.
24. Make the Simulation Reproducible
A communicated result should be sufficiently documented that another analyst can understand how it was generated.
Important details may include:
- Model version or release identifier.
- Parameter set or parameter distributions.
- Initial conditions.
- Dosing and intervention assumptions.
- Simulation duration and time step.
- Population-generation procedure.
- Random seed when stochastic simulation is used.
- Scenario definitions.
- Output definitions and transformations.
- Uncertainty and sensitivity-analysis methods.
These details do not all need to appear on the primary results slide. They can be provided in supplementary material, appendices, or technical documentation.
25. Common Mistakes When Communicating QSP Results
| Mistake | Why it causes problems | Better approach |
|---|---|---|
| Showing every model output | Important findings become difficult to identify | Select outputs tied directly to the scientific question |
| Hiding uncertainty | Creates an impression of false precision | Show relevant prediction or uncertainty ranges |
| Mixing observations and simulations | Blurs the evidentiary distinction | Use explicit visual and textual labels |
| Reporting a sensitivity ranking without context | Can be interpreted as universal biological importance | Specify output, parameter ranges, and analysis method |
| Presenting extrapolation as established fact | Hides dependence on model assumptions | Identify extrapolated conditions explicitly |
| Overloading figures with model detail | The scientific message becomes difficult to see | Move technical detail to supporting material |
| Using causal language too strongly | Can overstate what the model establishes | Use model-based and evidence-based language |
26. Connect the Result to the Decision
The final step in communicating a QSP result is to explain why the result matters. This does not mean turning a model prediction into a definitive decision. Instead, it means describing how the evidence informs the question that motivated the analysis.
For example, a simulation may indicate that:
- a particular dose range produces the desired modeled target engagement;
- additional target engagement produces diminishing incremental response;
- a prediction is highly sensitive to a particular biological parameter;
- an alternative mechanism produces substantially different outcomes;
- additional experimental data would meaningfully reduce uncertainty; or
- a proposed scenario falls outside the region where the model has strong supporting evidence.
These are useful scientific conclusions because they identify what the model adds to the evidence base without pretending that simulation alone resolves every uncertainty.
27. From a Complex Simulation to a Clear Scientific Message
Imagine that a QSP model contains 250 state variables, 180 parameters, several feedback loops, and simulations of multiple dose levels. The complete model may be technically complex, but the primary result may be summarized by a small number of questions.
| Communication question | Example answer |
|---|---|
| What was asked? | How does target inhibition affect the downstream disease biomarker? |
| What was simulated? | Three dose levels spanning low to high target engagement. |
| What was predicted? | Higher target engagement produced progressively lower modeled biomarker levels. |
| Why? | The model links target inhibition to suppression of a downstream biological pathway. |
| How certain? | The magnitude of the prediction varies across plausible parameter values. |
| What drives uncertainty? | The result is particularly sensitive to pathway turnover parameters. |
| What remains uncertain? | The model has limited evidence supporting extrapolation at very high target engagement. |
| What does the model contribute? | It provides a mechanistic framework for connecting target engagement to the downstream response. |
28. QSP Results Communication Checklist
Before finalizing a report, presentation, figure, or manuscript section, ask:
- Is the scientific question stated clearly?
- Is the purpose of the model explicit?
- Are observed data clearly distinguished from model predictions?
- Are treatment and scenario definitions unambiguous?
- Are the primary outputs directly relevant to the question?
- Are units and scales clearly labeled?
- Is uncertainty shown or described when it materially affects interpretation?
- Is the source of uncertainty identified?
- Are sensitivity results interpreted within the context of the analysis?
- Are mechanistic pathways explained rather than merely asserted?
- Are extrapolations identified?
- Are major model assumptions stated?
- Are limitations communicated explicitly?
- Can the simulation be reproduced from the documented inputs and settings?
- Does the conclusion distinguish model-based evidence from direct experimental evidence?
29. Key Takeaways
- Start with the scientific question. The purpose of QSP communication is to explain what the model contributes to a biological or development question.
- Separate observations from predictions. Model-generated trajectories and simulated outcomes should never be presented as though they were experimental measurements.
- Choose decision-relevant outputs. A complex QSP model can contain many variables, but only a subset usually needs to appear in the primary communication.
- Show uncertainty. A central prediction without its relevant uncertainty can create a misleading impression of precision.
- Explain mechanisms. QSP results are especially valuable when the predicted outcome can be connected to intermediate biological processes.
- Interpret sensitivity in context. An influential parameter for one output and parameter range is not automatically the most biologically important parameter in every context.
- Identify extrapolation. Predictions outside the range of supporting observations should be clearly identified as more dependent on model assumptions.
- Use precise language. Prefer "the model predicts" or "the model supports a mechanistic interpretation" when describing model-generated evidence.
- Make the analysis reproducible. Model version, parameters, scenarios, simulation settings, and output definitions should be documented even when they are not shown on the main slide.
- End with the scientific implication. The strongest QSP communication connects the model result back to the original question while clearly stating what remains uncertain.
Where to Go Next
A natural next step is to study QSP model qualification and credibility, including how model purpose, context of use, verification, validation, sensitivity analysis, uncertainty analysis, and external evidence support confidence in model predictions.
From there, related topics include model qualification for QSP, credibility assessment for mechanistic models, Bayesian calibration, parameter estimation, structural and practical identifiability, and sensitivity analysis.