Why Map Parameters Across Species?
A QSP model calibrated to several species can use all available measurements without pretending that every parameter is identical. A biological mechanism may be conserved while the quantities that describe it vary with body size, physiology, or species-specific baseline conditions. The modeling task is to state which parts of the model are shared, which are species-specific, and how the latter depend on the former.
A parameter map is a structural assumption, not merely a convenient way to combine estimates. If the map is biologically plausible and supported by the available information, it can reduce the number of independently fitted quantities and enable predictions for a species with limited data. If it is wrong, joint fitting can give precise-looking but biased estimates and predictions.
For observations \(y_{sj}\) from species \(s\), the model prediction can be written \(f_{sj}(\theta_s,\eta_s)\), where \(\eta_s\) collects any species-specific observation or nuisance parameters. With a proposed map, the prediction becomes \(f_{sj}(g_s(\phi,z_s),\eta_s)\). The data from the species then inform common parameters through their mapped predictions, while the model can still accommodate justified species differences.
Assumptions Behind a Cross-Species Map
A map is useful only if its scientific and mathematical assumptions are explicit. Before fitting, check that the following conditions are plausible:
- The represented mechanism is sufficiently conserved. The same equations or causal pathway must be a reasonable description in each species. A shared symbol does not establish shared biology.
- Species differences are represented by relevant inputs. Variables such as body weight or a measured physiological factor must be available and defined consistently. An unexplained species label is not a mechanistic covariate.
- Scales and units are compatible. Both sides of a map must have matching units, or the map must explicitly include a conversion factor with the appropriate units.
- The map form is justified independently of convenience. For example, an allometric exponent should be fixed from a defensible assumption or estimated only when the data can inform it. It should not be chosen solely because it makes the fit look better.
- The data contain information about mapped parameters. Different species’ data must collectively constrain the shared quantities; a map cannot make an uninformative data set informative by itself.
The earlier material on harmonizing units and time origins is especially relevant: a time-rate parameter reported per hour cannot be compared directly with one reported per day without conversion. The earlier material on joint calibration objectives also applies: observations contribute through their stated measurement and error models, not through arbitrary duplication of evidence.
Clearance Mapping by Body Size
A common illustrative map for clearance is an allometric relationship. For species \(s\), let \(W_s\) be body weight and let \(b\) be a chosen exponent. If \(CL_{\mathrm{ref}}\) is clearance for a reference weight \(W_{\mathrm{ref}}\), the map is
The weight ratio is dimensionless, so \(CL_s\) and \(CL_{\mathrm{ref}}\) have the same units. This equation does not say that clearance itself is conserved. It says that clearance is constrained to change with weight according to the selected exponent. The exponent and reference weight are part of the model specification, and other biological factors may make this simple relationship inadequate.
Worked Example: Mapping Clearance Estimates to a Shared Scale
Suppose hypothetical calibration estimates are available for a reference species and humans. After applying a prespecified weight-based map, the human-to-reference clearance factor is \(68.449\). The reference-species clearance estimate is \(0.620\ \mathrm{L/h}\) with standard error \(0.040\ \mathrm{L/h}\). The human estimate is \(42.0\ \mathrm{L/h}\) with standard error \(2.00\ \mathrm{L/h}\). Treat these as independent, approximately normal estimates, and assume the map factor is fixed and correct. Find a combined estimate on the reference scale.
First map the human estimate and its standard error to the reference scale by dividing each by the factor:
Under the stated independence and normal-approximation assumptions, the inverse-variance weights are the reciprocals of the squared standard errors:
The weighted estimate is the weighted sum divided by the sum of the weights. Using the displayed transformed estimate and weight gives
Mapping this result back to the human scale gives \(0.6158(68.449)\approx42.15\ \mathrm{L/h}\). This calculation is appropriate only under its stated simplifications: the estimates are independent, their standard errors adequately describe their uncertainty, and the map factor is treated as known. If the factor is uncertain, that uncertainty must also be represented rather than silently ignored.
Turnover: Shared Structure, Species-Specific Rates
Turnover models illustrate why mapping parameters is not the same as forcing all species to have the same numerical values. Consider a state \(R_s(t)\) with zero-input turnover described by
Here \(k_{\mathrm{in},s}\) has units of amount per time, and \(k_{\mathrm{out},s}\) has units of inverse time. At a positive steady state, the derivative is zero, so \(R_{0,s}=k_{\mathrm{in},s}/k_{\mathrm{out},s}\). Thus, the baseline level depends on both the input and loss rates. A shared turnover mechanism does not imply equal baselines or equal rates.
One possible map specifies \(k_{\mathrm{out},s}=g_s\kappa\), where \(\kappa\) is shared and \(g_s\) is a prespecified, dimensionless species factor. The input rate can remain species-specific, for example \(k_{\mathrm{in},s}=k_{\mathrm{out},s}R_{0,s}\) when baseline \(R_{0,s}\) is specified. This is a model choice: it shares a scaled turnover rate, not the raw rate or baseline state.
Worked Example: Mapping Turnover Rates With a Species Factor
In hypothetical data, estimated turnover half-times are \(3.466\) hours in species A and \(8.664\) hours in species B. Assume first-order turnover, so the half-time is \(t_{1/2}=\ln(2)/k_{\mathrm{out}}\). A prespecified map uses \(g_A=1\) and \(g_B=0.40\), with \(k_{\mathrm{out},s}=g_s\kappa\). Calculate the mapped estimates and predicted rates.
Converting each half-time to a rate gives
The implied shared-parameter estimates are \(\widehat{\kappa}_A=0.2000/1=0.2000\ \mathrm{h}^{-1}\) and \(\widehat{\kappa}_B=0.08000/0.40=0.2000\ \mathrm{h}^{-1}\). They agree with the proposed map: the mapped prediction for species B is \(0.40(0.2000)=0.08000\ \mathrm{h}^{-1}\). The different raw rates are therefore consistent with a common scaled rate in this example. That agreement alone does not establish that the map is biologically correct; it only shows the hypothetical values are compatible with it.
Deriving What the Map Implies
A short derivation clarifies what a turnover map predicts for both the time course and baseline. If \(k_{\mathrm{in},s}\) and \(k_{\mathrm{out},s}\) are constant, the steady state is \(R_{0,s}=k_{\mathrm{in},s}/k_{\mathrm{out},s}\). Write the deviation from baseline as \(D_s(t)=R_s(t)-R_{0,s}\). Since the derivative of the constant \(R_{0,s}\) is zero, substitution into the turnover equation yields
Multiplying by the integrating factor \(e^{k_{\mathrm{out},s}t}\) gives \(\frac{d}{dt}\left(e^{k_{\mathrm{out},s}t}D_s(t)\right)=0\), so \(D_s(t)=D_s(0)e^{-k_{\mathrm{out},s}t}\) for any initial value, including \(D_s(0)=0\). Therefore,
Under the map \(k_{\mathrm{out},s}=g_s\kappa\), the characteristic relaxation time is \(1/(g_s\kappa)\), and the half-time is \(\ln(2)/(g_s\kappa)\). These implications provide direct checks: the map predicts how quickly deviations from each species’ own baseline decay, not that the baselines must match.
Worked Example: Mapping Clearance and Volume Together
Suppose hypothetical rat parameters are \(CL_R=0.800\ \mathrm{L/h}\) and \(V_R=1.25\ \mathrm{L}\), at body weight \(4.00\ \mathrm{kg}\). A hypothetical human weighs \(64.0\ \mathrm{kg}\). Use \(CL\propto W^{0.75}\) and \(V\propto W^1\), with the rat values as reference. Calculate the mapped human clearance, volume, and elimination rate \(k=CL/V\).
The body-weight ratio is \(64.0/4.00=16.0\). The clearance factor is \(16^{0.75}=8.000\), and the volume factor is \(16.0\). Therefore,
The implied elimination rates are \(k_R=0.800/1.25=0.640\ \mathrm{h}^{-1}\) and \(k_H=6.40/20.0=0.320\ \mathrm{h}^{-1}\). Thus, even though both clearance and volume increase in the map, the elimination rate decreases by a factor of two. This illustrates why mapping one parameter in isolation can miss consequences for the model’s dynamics. The calculated values are model predictions conditional on the chosen exponents, not observed human measurements.
Choosing and Checking a Mapping Strategy
Three common strategies answer different questions. They should not be treated as interchangeable:
| Strategy | Constraint across species | When it may be useful | Main risk |
|---|---|---|---|
| Common raw parameter | The parameter has the same value in each species. | A mechanism is expected to be invariant on the parameter’s defined scale. | Real physiological differences are forced into other parameters or residual error. |
| Mapped parameter | Species values follow a specified function of shared parameters and species inputs. | There is a defensible relationship, such as a prespecified allometric scaling. | A wrong map can bias estimates and predictions across all species. |
| Independent parameters | Each species has its own value, without a cross-species constraint. | Species biology may differ and data are adequate to estimate those differences. | Parameters may be weakly informed, especially for sparse species data. |
A practical check is to calculate the implied parameter values for every species and inspect them in their original units and on the mapped scale. Then compare the model’s predictions with each species’ data in the same time and measurement units. A map that requires implausible values, produces systematic species-specific residual patterns, or fails to reproduce known baseline or time-scale differences warrants reconsideration. The earlier material on assay-specific observation error remains important: a systematic measurement-scale difference should not automatically be attributed to biology.
If the data are sparse, multiple maps may produce nearly indistinguishable predictions. In that case, the data may not identify the map form or its exponent. A prespecified biological constraint, additional informative measurements, or a clearly stated uncertainty range may be more defensible than estimating an elaborate species relationship from insufficient information. A good fit is necessary but does not prove that a map is mechanistically true.
Common Pitfalls and Limitations
- Equating conserved mechanism with identical values. The same turnover equation can apply while its rates and baselines differ. State exactly what is shared.
- Comparing values with incompatible units. Convert units before interpreting a rate or applying a map; a factor of 24 is introduced when converting a per-day rate to a per-hour rate.
- Ignoring uncertainty in map inputs. Treating an estimated scaling factor, body weight, or exponent as fixed can understate uncertainty in mapped predictions.
- Assuming all species follow one relationship. A map may omit species-specific biology, or the mechanism may not be conserved over the relevant range. Inspect species-wise predictions and residuals.
- Overinterpreting a shared estimate. A shared parameter is a consequence of the specified model structure; it is not proof that the underlying biological quantity is literally identical.
Parameter mapping is most valuable when it makes a testable biological statement: given specified species characteristics, the model predicts particular parameter values and thus particular dynamics. Its limitations are equally important. Predictions for a new species require extrapolation of the assumed relationship, and the reliability of that extrapolation depends on how well the map is supported by biological knowledge and the observed species range.
Check Your Understanding
Answer each question, then compare with the solutions below.
- A model uses \(CL_s=CL_{\mathrm{ref}}(W_s/W_{\mathrm{ref}})^{0.75}\). What is shared, and what is species-specific?
- If \(k_{\mathrm{out}}=0.10\ \mathrm{h}^{-1}\), what is the first-order turnover half-time to three significant figures?
- Two species have the same turnover rate but different \(k_{\mathrm{in}}\). Must their steady-state levels be equal? Explain.
- Give one reason independent species-specific parameters might be preferable to a mapped parameter, and one cost of that choice.
Solutions. (1) The reference clearance and the specified exponent are shared model quantities; each species’ body weight, and hence its mapped clearance, is species-specific. (2) \(t_{1/2}=\ln(2)/0.10=6.93\ \mathrm{h}\). (3) No. Since \(R_0=k_{\mathrm{in}}/k_{\mathrm{out}}\), different input rates with the same loss rate give different steady states. (4) Independent values may be preferable when there is no defensible common mapping or species biology differs. The cost is that each species’ parameter requires information from its own data and may be poorly estimated when those data are sparse.