1. What Are the Building Blocks of a QSP Model?
Quantitative systems pharmacology (QSP) models describe biological systems using mathematical representations of mechanisms that connect disease biology, drug exposure, pharmacology, and clinical outcomes.
A QSP model is typically assembled from several different types of objects. Some represent quantities that change over time, some represent fixed or slowly varying characteristics of the system, and others represent external inputs or measurable outputs.
A QSP model combines biological components, dynamic states, parameters, inputs, and mechanistic relationships to generate predictions.
```2. What Is a Model Component?
A model component is a conceptual element of the biological system represented in the model. Components can represent entities such as cells, tissues, signaling molecules, cytokines, receptors, disease processes, drug molecules, or physiological processes.
The precise meaning of a component depends on the model. A component may correspond relatively directly to a biological entity, or it may represent an abstraction that summarizes several biological processes.
| Component type | Example | Role in a QSP model |
|---|---|---|
| Drug | Unbound drug concentration | Represents drug exposure available to interact with biological targets. |
| Target | Receptor or enzyme | Represents a molecular target through which drug action may occur. |
| Cell population | Tumor cells | Represents a population whose abundance may change over time. |
| Biomarker | Circulating cytokine | Represents a measurable biological quantity linked to mechanism or response. |
| Physiological process | Cell proliferation | Represents a rate or mechanism that changes one or more states. |
| Disease process | Disease progression | Represents changes in disease burden or disease-related biology. |
Components therefore provide the conceptual vocabulary of a QSP model. They tell the modeler what biological entities and processes are being represented before the mathematical relationships among them are fully specified.
3. What Is a State Variable?
A state variable is a quantity whose value can change as the system evolves. In a dynamic QSP model, states are usually functions of time.
Examples include the amount of drug in a compartment, the number of tumor cells, the concentration of a cytokine, the fraction of occupied receptors, or the number of activated immune cells.
If \(X(t)\) represents a biological state, then its value at time \(t\) summarizes some aspect of the system at that moment.
The defining feature is that the state can evolve. Its change is determined by the processes represented in the model.
4. States and Parameters Are Not the Same
One of the most important distinctions in mechanistic modeling is the difference between states and parameters.
A state generally changes during a simulation. A parameter generally defines a property of the model and remains fixed during a particular simulation unless the model explicitly allows it to vary.
| Feature | State | Parameter |
|---|---|---|
| Typical behavior | Changes over time | Usually fixed during a simulation |
| Example | Tumor cell count | Tumor growth rate |
| Mathematical role | Appears as a dynamic variable | Controls the dynamics |
| Initial value | Usually required | Usually specified or estimated |
| Typical interpretation | Current system condition | Property controlling system behavior |
For example, consider a simple tumor-growth model:
Here \(T(t)\) is a state representing tumor burden, while \(r\) is a parameter representing the intrinsic growth rate. The state changes as time progresses; the parameter controls how rapidly that change occurs.
5. What Are Parameters in QSP?
Parameters are quantities that determine the behavior of the mathematical model. They may represent biological rates, binding affinities, concentrations, capacities, thresholds, delays, or other characteristics of the modeled system.
Parameters often have direct mechanistic interpretations. For example, a receptor-binding model may contain an association rate constant, dissociation rate constant, and receptor abundance. A cell-growth model may contain proliferation and death rates.
| Parameter | Possible interpretation | Example role |
|---|---|---|
| \(k_{\mathrm{on}}\) | Association rate | Controls formation of a drug-target complex. |
| \(k_{\mathrm{off}}\) | Dissociation rate | Controls loss of a drug-target complex. |
| \(K_D\) | Binding affinity parameter | Characterizes equilibrium binding behavior. |
| \(r\) | Growth rate | Controls expansion of a modeled cell population. |
| \(k_{\mathrm{death}}\) | Death rate | Controls removal of cells from a population. |
| \(EC_{50}\) | Concentration producing half-maximal effect in an appropriate model | Controls the concentration-effect relationship. |
Parameters are therefore the quantities that give biological meaning and scale to the model equations.
6. Initial Conditions: Where Does the Model Start?
A dynamic model needs to specify the starting value of its states. These are called initial conditions.
If a model contains \(n\) states, it can be written generally as:
where \(\mathbf{x}\) is the vector of states and \(\boldsymbol{\theta}\) is the vector of parameters. The initial condition is:
The initial condition can be biologically important. For example, a disease model may start with a baseline disease burden, while an immune-response model may start with baseline immune-cell populations and biomarker concentrations.
7. How States and Parameters Become Equations
States and parameters become useful when they are connected through equations that describe biological processes.
Suppose \(T(t)\) represents tumor burden. If tumor cells proliferate at rate \(r\) and are killed at rate \(k_{\mathrm{kill}}\), a simplified model might be:
The first term increases tumor burden, while the second term decreases it.
The equation can be expanded to represent drug-mediated killing:
Here the killing rate depends on drug concentration \(C\). The model therefore creates a mechanistic connection between exposure and disease biology.
8. Rate Laws Describe How States Change
Most dynamic QSP models are constructed from rate laws. A rate law specifies how quickly a process changes one or more states.
For a simple first-order process:
The rate of disappearance is proportional to the current amount of \(X\).
Biological systems frequently require more structured rate laws. For example, receptor-mediated binding can be represented using:
where the terms describe formation and dissociation of a bound complex under the particular definitions of the states.
Other common QSP rate laws include Michaelis-Menten kinetics, Emax relationships, logistic growth, Hill functions, turnover models, receptor occupancy models, and target-mediated drug disposition relationships.
9. Components Become a Biological Network
QSP models often contain many interacting components. Rather than modeling each component independently, the model connects them into a network of biological relationships.
A simplified QSP network can connect drug exposure to target engagement, downstream signaling, cellular behavior, and measurable biomarkers.
```The network perspective is central to QSP. A drug may affect a target, which changes a signaling pathway, which changes a cellular state, which eventually changes a disease biomarker or clinical endpoint.
Each arrow in such a diagram should correspond to an explicit mathematical relationship in the model rather than merely representing a conceptual association.
10. States Are Not Always Directly Observed
An important feature of QSP models is that many biologically important states cannot be measured directly in a clinical study.
For example, a model may contain a state representing intracellular pathway activity even though the clinical dataset contains only a circulating biomarker. The model must therefore connect latent states to observable quantities.
A simple observation relationship might be:
where \(X(t)\) represents a model state, \(Y(t)\) represents an observed quantity, \(h(\cdot)\) is an observation function, and \(\varepsilon(t)\) represents measurement or residual variability.
| Model quantity | Possible clinical observation |
|---|---|
| Target occupancy | Imaging or pharmacodynamic biomarker |
| Activated immune-cell population | Peripheral blood cell measurement |
| Tumor burden | Tumor size or imaging-derived measurement |
| Pathway activity | Downstream biomarker |
| Drug concentration | Measured plasma or tissue concentration |
This distinction between model states and observations is critical. A model can contain biologically meaningful quantities that are inferred indirectly rather than measured directly.
11. Different Types of QSP Parameters
Not all parameters play the same role. Classifying them helps clarify what information is coming from experiments, literature, clinical data, or model assumptions.
| Parameter category | Example | Typical source |
|---|---|---|
| Drug-specific | Binding affinity | In vitro or translational pharmacology studies |
| Physiological | Organ volume or blood flow | Physiology literature or clinical data |
| Disease-specific | Baseline disease burden | Clinical or natural-history data |
| Pharmacodynamic | Maximum drug effect | In vitro, animal, or clinical exposure-response data |
| Population variability | Between-subject variance | Population data or hierarchical modeling |
| Empirical or calibration | Scaling coefficient | Model calibration against observations |
A QSP model can therefore integrate information from multiple experimental and clinical sources. One purpose of the model is to make those assumptions explicit and connect them into a coherent mechanistic framework.
12. Why Units Matter in QSP Models
Every state, parameter, input, and equation in a quantitative model should have internally consistent units.
Suppose a state \(X\) represents a concentration measured in mg/L and a first-order rate constant \(k\) is measured in \(1/\mathrm{h}\). Then:
has units of:
Dimensional consistency provides a basic but powerful model-checking mechanism. Unit errors can otherwise propagate through an entire QSP model and produce apparently plausible but quantitatively incorrect predictions.
13. Worked Example: A Simple Drug-Target Model
Consider a simplified QSP model describing binding between a drug \(D\) and a biological target \(R\). The drug-target complex is \(DR\).
The conceptual mechanism is:
Step 1: Define the states
Suppose the model tracks free drug concentration \(D(t)\), free target concentration \(R(t)\), and drug-target complex concentration \(DR(t)\).
Step 2: Define the parameters
Let \(k_{\mathrm{on}}\) describe association and \(k_{\mathrm{off}}\) describe dissociation.
Step 3: Define the binding rate
The association rate can be represented as:
Here the notation \(DR\) in the product should be interpreted carefully in implementation: if \(D\) and \(R\) denote concentrations, the association term is \(k_{\mathrm{on}}D R\), while \(DR\) denotes the complex.
Step 4: Define the dissociation rate
Step 5: Write the complex equation
Step 6: Interpret the model
Increasing \(k_{\mathrm{on}}\) makes complex formation faster, while increasing \(k_{\mathrm{off}}\) makes dissociation faster. The state \(DR(t)\) changes over time according to the balance between these competing processes.
This small example contains the essential ingredients of a QSP model: components (drug and target), states (their modeled quantities), parameters (rate constants), and equations connecting them mechanistically.
14. Inputs Are Different From States and Parameters
QSP models also require inputs. Inputs represent quantities imposed on the system rather than generated solely by the model's internal dynamics.
A drug dose is a common example. In a dosing simulation, the dose enters the system according to a specified schedule.
This expression is one idealized way to represent bolus doses occurring at times \(t_j\). Other implementations may represent oral absorption, infusion rates, or repeated dosing using different input functions.
| Object | Example | Question it answers |
|---|---|---|
| State | Tumor burden | What is the current condition of the system? |
| Parameter | Tumor growth rate | What rule controls system behavior? |
| Input | Drug dose | What external intervention is applied? |
| Output | Tumor size | What does the model predict or compare with data? |
15. How Are QSP Parameters Determined?
QSP parameters can come from many sources. Some are measured directly in experiments, some are obtained from published literature, and others are estimated or calibrated using data.
- Direct measurement. Some quantities can be measured experimentally and incorporated directly.
- Literature values. Published physiological or pharmacological measurements can provide prior information.
- Scaling. Experimental information may need to be translated between species, tissues, or experimental systems.
- Parameter estimation. Parameters can be estimated by fitting the model to observed data.
- Calibration. Some parameters may be adjusted so that the model reproduces selected reference observations.
The distinction between measured and estimated parameters should be documented. A parameter that has been calibrated to one dataset may carry different uncertainty and interpretation than a parameter measured independently in a controlled experiment.
16. Can the Data Identify Every Parameter?
A model can contain more parameters than the available data can reliably inform. This creates an important issue known as identifiability.
Suppose two parameters always occur in an equation as a product:
If the available observations only reveal the product \(ab\), the data may not contain enough information to determine \(a\) and \(b\) separately.
QSP models can be particularly vulnerable to this issue because they may contain many mechanistic parameters while clinical datasets contain relatively few observations.
| Question | Why it matters |
|---|---|
| Is the parameter structurally identifiable? | Determines whether the mathematical model permits unique recovery of the parameter from ideal observations. |
| Is the parameter practically identifiable? | Determines whether the available noisy data contain enough information to estimate it precisely. |
| Does the parameter have external information? | Prior experimental or literature information may constrain otherwise weakly informed parameters. |
| Does the parameter affect the decision or prediction? | Some poorly estimated parameters may have limited impact on the scientific question. |
Identifiability is therefore not merely a statistical technicality. It influences how much mechanistic detail can be supported by the available evidence.
17. Building a Reliable QSP Model
Before using a QSP model for simulation or decision support, the model should be checked at multiple levels.
- Conceptual verification. Confirm that the model represents the intended biological mechanisms and assumptions.
- Implementation verification. Confirm that the computer implementation correctly represents the mathematical equations.
- Unit checking. Verify that equations are dimensionally consistent.
- Limiting-case testing. Check whether the model behaves sensibly when parameters or inputs approach important limiting values.
- Numerical testing. Evaluate solver behavior and numerical stability under relevant simulation conditions.
- Model evaluation. Compare model predictions with appropriate experimental or clinical observations.
- Sensitivity analysis. Determine which parameters and assumptions have meaningful effects on the outputs of interest.
18. A Practical Workflow for Defining QSP Model Components
- Start with the biological question. Define what mechanism, treatment response, or disease process the model needs to represent.
- Identify the important components. List the drug, targets, cell populations, biomarkers, physiological processes, and disease mechanisms relevant to the question.
- Define the states. Determine which quantities need to change dynamically during simulation.
- Define the parameters. Specify the biological properties and rates that govern the system.
- Define initial conditions. Specify the starting state of the biological system.
- Define inputs and interventions. Specify doses, treatment schedules, environmental changes, or other external drivers.
- Write the mechanistic equations. Translate biological assumptions into mathematical relationships.
- Define observables. Specify how model states are connected to measurable experimental or clinical quantities.
- Check units and implementation. Verify dimensional consistency and computational implementation.
- Assess identifiability and sensitivity. Determine what the available evidence can actually support.
- Evaluate model predictions. Compare predictions with appropriate observations and document limitations.
19. Key Takeaways
- QSP models represent biological systems using interconnected components, states, parameters, inputs, outputs, and mechanistic equations.
- Components define the biological entities and processes represented in the model.
- States are dynamic quantities that describe the condition of the system and can change over time.
- Parameters describe properties or rates that control how states change.
- Initial conditions specify where the dynamic system starts.
- Inputs such as drug doses represent external interventions rather than internally generated states.
- Rate laws translate biological mechanisms into mathematical relationships between states and parameters.
- Model states do not necessarily correspond directly to measurable clinical observations; observation functions can connect latent states to data.
- Units and dimensional consistency are essential for implementing quantitative mechanistic models correctly.
- Parameter provenance matters because QSP parameters may come from direct measurements, literature, scaling, estimation, or calibration.
- Identifiability determines whether available information can distinguish among parameters and mechanisms.
- Verification, sensitivity analysis, and model evaluation are essential before relying on QSP predictions.
- A well-designed QSP model represents enough biological detail to address its scientific question without adding unsupported complexity.
Where to Go Next
A natural next step is to learn how these building blocks are assembled into a complete QSP model. This includes defining the biological scope, drawing a mechanism map, identifying state variables, writing conservation and turnover equations, specifying drug mechanisms, and connecting model states to biomarkers and clinical outcomes.
From there, QSP modeling can progress to model scope and boundary definition, building a QSP model from biological knowledge, mechanistic drug-action models, parameterization, sensitivity analysis, uncertainty analysis, model qualification, and translational simulation.
References
- Sorger PK, Allerheiligen SRB, Abernethy DR, et al. Quantitative and Systems Pharmacology in the Post-genomic Era: New Approaches to Discovering Drugs and Understanding Therapeutic Mechanisms. NIH Quantitative and Systems Pharmacology Workshop report.
- Koch G, Schropp J, Jusko WJ. Assessment of nonlinear pharmacokinetic and pharmacodynamic models in systems pharmacology.
- Gadkar KG, Fillicori F, Feng J, et al. A framework for quantitative systems pharmacology modeling and model qualification.
- European Medicines Agency. Guideline on the qualification and reporting of physiologically based pharmacokinetic (PBPK) models and related model-informed drug development approaches.
- U.S. Food and Drug Administration. Model-Informed Drug Development: Frequently Asked Questions.
QSP terminology varies somewhat across modeling frameworks and software environments. The definitions used in this tutorial are intended as general modeling concepts rather than requirements of a particular QSP platform.