Why Numerical Agreement Is Not Enough
Two papers may report values for the same named protein or drug, yet those values need not represent the same quantity. One enzyme-abundance estimate might come from healthy adult liver microsomes and be normalized per milligram of microsomal protein; another might describe protein per gram of whole liver in a diseased population. Their numbers cannot be compared directly just because both refer to the same enzyme. The same issue arises for plasma binding: values can depend on the biological source, drug concentration, assay conditions, and the precise binding endpoint.
The practical question for PBPK input selection is therefore not simply, “Which number is larger?” First ask what was measured, in whom or in what material, under which conditions, and how the endpoint was defined. The earlier material on reconstructing assay metadata provides the publication-level details needed for this assessment; here the focus is how those details determine whether two results are comparable.
A Practical Test for Comparability
Treat each reported value as an estimate produced by a particular measurement system, not as a context-free property. A useful conceptual representation is
Here \(y_i\) is the reported result from measurement \(i\); \(\theta\) denotes the underlying biological property of interest; \(P_i\) is the population or biological source; \(C_i\) is the assay condition; \(H_i\) is sample handling; \(E_i\) is the endpoint definition; \(N_i\) is the normalization basis; \(g_i\) describes how those features map the underlying property to the reported scale; and \(\varepsilon_i\) represents measurement error. This is a conceptual model, not a claim that all effects can be separated or estimated from a publication.
Direct comparison is most defensible when the two measurements target the same property, relevant populations and conditions are aligned, and endpoint and normalization definitions agree. Differences in instrument or assay platform do not automatically make results incomparable: they may be comparable if both methods measure the same endpoint and their agreement has been established for relevant samples. Conversely, the same instrument does not guarantee comparability if the samples or endpoints differ.
When a feature differs, ask whether it changes the quantity being estimated or only the way it is expressed. A change of units or a valid, documented normalization conversion may preserve the underlying quantity. A change from protein abundance to catalytic activity does not: activity also depends on enzyme function and assay conditions. Likewise, a difference in population may reflect genuine biology rather than a mere scale difference. In those cases, numerical conversion alone cannot make the measurements equivalent.
Work Through the Measurement Definition
State what the result represents, such as amount of a specified enzyme protein or unbound fraction of a specified drug. Do not rely on a broad label such as “CYP value” or “binding.”
Check tissue or matrix, species, relevant demographic or disease features, and any treatment or induction context that could change the measured property.
Check assay temperature, pH, concentration, incubation or equilibration time, storage, freeze-thaw history, and other conditions material to the endpoint.
Determine what was counted or calculated, how it was normalized, and whether an evidenced conversion is available. If a key definition cannot be recovered, record the uncertainty rather than presuming equivalence.
These steps are a reasoning sequence, not a numerical score. Strong agreement on some features does not cancel a mismatch in a feature that changes the measurand. For example, identical units cannot compensate for comparing enzyme abundance in one sample with enzyme activity in another.
Worked Example: Enzyme Abundance Versus Enzyme Activity
Suppose two hypothetical reports concern a liver enzyme. Report A measures immunoreactive enzyme protein in healthy-adult liver microsomes and gives 85 pmol per milligram of microsomal protein. Report B measures substrate turnover in liver microsomes and gives 0.42 nmol per minute per milligram of microsomal protein. Can the values be compared as two estimates of enzyme abundance?
No. Report A’s endpoint is the amount of detected protein per mass of microsomal protein. Report B’s endpoint is catalytic activity per mass of microsomal protein. Although both use microsomes and a similar denominator, their numerators have different meanings and units: pmol of protein versus nmol of product formed per minute.
The difference is not repaired by converting units. For example, converting Report A’s numerator from pmol to nmol gives \(85\ \text{pmol/mg} = 0.085\ \text{nmol/mg}\), but that is still an amount of protein, not an amount of product formed per minute. Activity could be related to enzyme amount only with additional information and an appropriate model of catalytic function under the assay conditions. The reports therefore do not directly measure the same measurand.
For a PBPK input specifically requiring enzyme abundance, Report A is the relevant type of measurement, subject to whether its population and assay apply to the intended model. Report B may inform a functional activity parameter, but it cannot be substituted as an abundance estimate on the basis of these data alone.
Normalization and Biological Context
A reported value is often a ratio. The numerator and denominator both matter: “enzyme abundance” could mean protein amount per milligram of microsomal protein, per gram of liver, or per whole organ. These are distinct scales. A conversion is defensible only if the additional quantity needed for the conversion is appropriate to the relevant source and the definitions are compatible. A value measured in a different population may still be useful, but the population difference is an applicability question, not a unit conversion.
Worked Example: Converting Enzyme Abundance to a Common Scale
Consider two hypothetical measurements intended to describe the same enzyme protein in comparable healthy-adult liver material. Study A reports 120 pmol per milligram of microsomal protein. A separately measured, relevant microsomal-protein yield is 35 mg per gram of liver. Study B reports 4.0 nmol per gram of liver. Determine whether the values agree after expressing them on the same scale.
Using the yield, convert Study A’s result to a per-gram-of-liver scale:
Since \(1000\ \text{pmol} = 1\ \text{nmol}\), this is \(4.2\ \text{nmol/g liver}\). Study B’s value is \(4.0\ \text{nmol/g liver}\), a difference of \(0.2\ \text{nmol/g liver}\), or 5% of Study B’s value. The converted values are numerically close.
That closeness is meaningful only if the yield applies to the relevant biological material and the measurements target the same enzyme protein, population, and compatible endpoints. If the yield came from another population or an incompatible preparation, the calculation is arithmetically valid but may not establish comparability. Even with aligned definitions, a 5% difference should not be called statistically insignificant without information about measurement uncertainty and study design. The warranted conclusion is narrower: under the stated assumptions, the values are on a common scale and are numerically similar.
Plasma Binding: Define the Endpoint and Conditions
For a drug in plasma, the unbound fraction is the concentration of unbound drug divided by the total drug concentration in the specified sample and under specified conditions. The bound fraction is a different endpoint: if the measured drug is partitioned only into unbound and bound forms, then bound fraction equals one minus unbound fraction. The endpoint, drug concentration, matrix, and assay conditions must be checked before comparing results.
Assay method and handling matter. Equilibrium dialysis and ultrafiltration, for example, use different procedures to separate unbound from protein-associated drug. Results from these methods are not automatically interchangeable; evidence is needed that they give compatible estimates for the relevant drug, sample, and conditions. Temperature, pH, drug concentration, dilution, equilibration time, nonspecific binding to equipment, and sample storage can affect the measured result or its interpretation. Serum and plasma are not interchangeable labels: their compositions differ, so a measurement in one does not automatically represent the other.
Worked Example: Comparing Plasma Unbound Fractions
Two hypothetical reports estimate binding of the same drug in human plasma. Report A gives a bound fraction of 98.2% using equilibrium dialysis at 37°C and a total drug concentration of 1.0 mg/L. Report B gives \(f_u = 0.025\) using ultrafiltration at 25°C and a concentration of 10 mg/L. Both used stored plasma, but the storage details and nonspecific-binding checks are not available. Are the numbers directly comparable?
First express the endpoints in the same form. Under the stated two-fraction definition, Report A implies
Report B’s \(f_u\) is 0.025. The difference is \(0.025 - 0.018 = 0.007\), or 0.7 percentage points. That calculation establishes that the reported endpoints can be expressed on a common scale; it does not establish that they estimate the same condition-specific quantity.
The methods, temperature, concentration, and incompletely described handling differ. Because binding may depend on concentration and assay conditions, and because method-specific effects have not been ruled out, the values are not directly comparable as estimates under a single shared condition. They may be treated as separate, condition-dependent evidence, but selecting one as interchangeable with the other would require further support, such as method agreement under matched conditions. The arithmetic difference alone does not show which report is closer to an intended in vivo value.
Justification: Why the Definition Comes Before the Difference
A simple derivation clarifies the logic. Suppose two reported values are genuinely on the same scale and target one common quantity \(\theta\), with measurement errors \(e_1\) and \(e_2\). Then
Subtracting gives \(y_1-y_2=e_1-e_2\). In this case, their difference reflects measurement error and any sampling variation represented by the errors. But if the measurements target different condition-specific quantities, write \(y_1=\theta_1+e_1\) and \(y_2=\theta_2+e_2\). Then
Now the difference combines a genuine difference in the quantities with measurement error. Without evidence connecting \(\theta_1\) and \(\theta_2\), the observed difference cannot be interpreted as assay disagreement alone. This is why a numerical difference—or numerical agreement—cannot by itself establish comparability.
Diagnostics, Alternatives, and Limits
Look for warning signs before treating values as interchangeable: different tissue or matrix, populations with plausible biological differences, activity reported where abundance is needed, inconsistent denominators, distinct concentration or temperature conditions, and missing handling or endpoint definitions. Close numerical values are not a diagnostic that resolves these issues. A large discrepancy does not prove the measurements are incomparable either; it may be a real biological difference, a method effect, or both.
There are three practical alternatives when definitions do not align. Use the measurement that directly matches the intended PBPK input; keep the results separate as evidence for different populations or conditions; or seek a justified conversion or method-bridging relationship. The last option needs evidence appropriate to the mismatch. Unit conversion resolves units, not biological differences. A conversion between normalization bases needs a relevant yield or denominator. A method bridge needs evidence that the methods estimate compatible endpoints under relevant conditions.
Comparability is not the same as representativeness. Two assays may measure the same quantity in the same way but use a population that is not suitable for the intended model. Conversely, two measurements may be useful for understanding the same biology while remaining non-interchangeable. Assessment is also limited by what the source reports: unknown sample handling or an ambiguous endpoint should remain an explicit limitation, not be silently filled in by assumption.
Common Mistakes
- “The units match, so the measurements are comparable.” Matching units does not align analyte, population, condition, endpoint, or denominator.
- “The values are close, so the methods agree.” Numerical closeness is not evidence of method agreement when the conditions or measurands differ.
- “Bound fraction and unbound fraction are both binding, so either number can be used.” They are complementary endpoints under a specified two-fraction definition; convert them explicitly and check that definition.
- “A per-milligram value can be compared with a per-gram value after changing units.” Different biological denominators require a relevant normalization quantity, not just a metric prefix conversion.
- “An enzyme activity value is an enzyme abundance value in different units.” Activity and protein amount are distinct properties and cannot be equated without additional evidence.
Check Your Understanding
Answer each question, then check the solutions below.
- One study reports enzyme protein in pmol per milligram of microsomal protein; another reports enzyme activity in nmol per minute per milligram. Are the endpoints directly comparable? Why?
- A report gives \(f_u=0.03\). Under the two-fraction definition, what bound fraction does this imply?
- What additional quantity is needed to convert enzyme abundance per milligram of microsomal protein to abundance per gram of liver?
- Two plasma-binding assays use different temperatures and drug concentrations but report similar \(f_u\). What can and cannot be concluded from that similarity?
- Give one example of a difference that may be resolved by a justified conversion and one that cannot be resolved by unit conversion alone.
Solutions. (1) No. One endpoint is protein amount and the other is catalytic activity; the matching denominator does not make their numerators equivalent. (2) \(f_{\mathrm{bound}}=1-0.03=0.97\), or 97%. (3) A relevant microsomal-protein yield, expressed as milligrams of microsomal protein per gram of liver. (4) Their results are numerically similar on the reported scale, but the condition differences mean they have not been shown to estimate the same condition-specific quantity. (5) A change from pmol to nmol can be resolved by unit conversion. A change from protein abundance to catalytic activity cannot be resolved by unit conversion alone.