HealthyWelness
Account Cart0

Accuracy vs Precision in Peptide Analysis: Bias, Trueness and Uncertainty

Distinguish analytical precision, trueness, bias and uncertainty, with hypothetical comparisons and clear limits on repeatability claims.

AI-generated conceptual editorial artwork: Scientist examining a cuvette under soft window light, warm neutral palette. Not exact molecular structure, actual staff or HealthyWelness facility.

Precision concerns agreement among repeated measurements under stated conditions; accuracy concerns closeness to the value of the defined measurand. A tightly clustered set of peptide measurements can still be systematically displaced. Trueness, estimated bias, precision and measurement uncertainty help describe different parts of that problem and should not be replaced by a single unexplained “accuracy percentage.”

Start with what the measurement is supposed to represent

Before deciding whether a peptide result is accurate, identify the quantity being measured. Target-peptide mass fraction in a powder, concentration in an analysed solution, chromatographic area fraction and an instrument’s mass-to-charge indication are different measurands. A good result for one does not automatically establish another.

For example, two repeated concentration estimates can agree closely without establishing the target sequence. Conversely, evidence supporting identity does not alone establish the amount present. The analytical question should therefore appear alongside its units, molecular identity, sample basis and method context.

The International Vocabulary of Metrology (VIM) treats accuracy as a qualitative concept associated with closeness to a true quantity value. It explicitly distinguishes accuracy from precision and trueness. That vocabulary discourages giving “accuracy” an undefined numerical value as though it were a separate physical quantity. [1]

A useful reading habit is to translate “accurate” into a request for evidence: agreement with what reference, over what range, in what sample context and with what uncertainty? This makes the statement reviewable instead of leaving its meaning to the reader’s intuition.

Trueness and bias address the centre of repeated results

VIM defines trueness in terms of the average that would result from indefinitely many replicate measurements compared with a reference value. It relates trueness to systematic effects rather than random dispersion. A finite set of results cannot reveal that infinite average exactly; it supplies evidence from which a displacement may be estimated. [2]

In a simple comparison, estimated bias is the measured mean minus the reference value, with the sign retained. A positive difference means the reported mean is higher on the chosen scale; a negative difference means it is lower. A relative difference divides by the stated reference value and therefore needs that denominator to be explicit.

Such a comparison is only as meaningful as its reference and design. A reference value also has uncertainty. A small observed difference may be compatible with uncertainty on both sides, while a large difference may suggest a problem that cannot be explained by repeatability alone. Neither observation identifies the cause without further evidence.

A report should not simply relabel the mean of a few unknown samples as the “true value.” Nor should an agreement with one assigned value be silently expanded into a claim about every concentration, matrix or preparation the method might encounter.

Precision needs a description of the conditions

VIM describes precision through agreement among replicate measurements under specified conditions, commonly summarised by SD, variance or coefficient of variation (CV). The conditions are part of the statement: a precision statistic without them leaves unclear which sources of variation were allowed to contribute. [3]

Different precision claims include different variation
Reported comparison What it can examine What remains outside that comparison
Repeated readings of one preparation Variation in the repeated measurement stage Independent preparation errors and broader sample variation
Independent preparations under repeatability conditions Preparation and measurement variation within those conditions Changes of day, operator or location that were not included
Intermediate precision Specified changes over time within the same location and procedure Unexamined changes and general agreement between laboratories
Reproducibility comparison Agreement under explicitly changed conditions, potentially across locations Universal portability to every laboratory and sample

NIST’s terminology guidance describes repeatability under the same conditions over a short period and stresses specifying changed conditions when discussing reproducibility. These labels describe the scope of the comparison, not a ranking of laboratories. [4]

The current VIM entry for intermediate precision specifically retains the same procedure and location over an extended period while allowing specified changes, such as operators, calibrations or measuring systems. A report should identify what changed and what stayed fixed. [5]

For a peptide analysis, six repeated injections from one vial are therefore a different evidence statement from six independently prepared sample solutions. Both may be useful, but they should not be counted as equivalent investigations of the whole analytical process.

Worked example: a tight cluster can be displaced

Invented summary comparison: imagine an appropriate reference material with an assigned concentration of 100.0 units. The table gives hypothetical method means and sample SDs from comparable, independently prepared analytical portions. The reference’s uncertainty is deliberately not supplied, so the example illustrates arithmetic rather than a formal compatibility assessment.

Hypothetical summaries, not raw observations or product results
Method summary Mean SD Difference from reference CV
A 108.0 1.0 +8.0 units; +8.0% 1.0 / 108.0 × 100 ≈ 0.93%
B 100.5 4.0 +0.5 units; +0.5% 4.0 / 100.5 × 100 ≈ 3.98%
C 100.5 1.0 +0.5 units; +0.5% 1.0 / 100.5 × 100 ≈ 1.00%

A has less observed dispersion than B, yet its mean is farther from the assigned reference. B’s mean is closer but its measurements are more dispersed. C combines a closer mean with lower dispersion in this invented comparison. None of these rows provides a universal pass or fail threshold.

The CV uses each row’s mean as its denominator, while the relative difference uses the reference value. That is why the denominators differ. A CV of 0.93% is not evidence that the estimate is within 0.93% of the reference; it describes relative dispersion within A’s reported results.

These summaries cannot supply a confidence interval without the relevant sample size and design, and they cannot establish the true long-run bias without qualification. No raw observations are invented to fill those gaps. The example simply separates a centre comparison from a dispersion comparison.

More repeats can sharpen the wrong answer

Suppose a shared conversion factor systematically shifts every result. Repeating the same calculation can reduce uncertainty about the average indication without revealing that factor. A small SEM may then coexist with a displaced mean. Additional readings address sampling variation within their scope, not every shared assumption.

This is why repeatability evidence and reference comparisons are complementary. The SD, SEM and confidence-interval guide explains why a precise mean is different from tightly dispersed observations. The calibration and reference-standard guide covers the separate relationship between the response scale and assigned analyte levels.

Reading those questions separately also prevents a common shortcut: treating a good calibration fit as independent evidence of accuracy. A shared assigned-value or preparation error can remain shared throughout the fitted relationship. The issue here is the dependency, not whether a line looks straight.

Uncertainty is broader than repeatability

Measurement uncertainty describes the dispersion of values attributable to the measurand using the available information. A useful evaluation may include reference assignment, preparation, calibration, correction and measurement components, with their relationships represented in the model. It should not automatically be replaced by the SD from repeat readings.

NIST distinguishes Type A evaluation by statistical methods from Type B evaluation by other information. It also cautions that these categories do not simply correspond to “random” and “systematic,” and that uncertainty is distinct from error. [6]

For a hypothetical interpretation, a result may have small repeated-reading SD while the assigned reference value is relatively uncertain. Ignoring the latter would make the displayed uncertainty look smaller without improving the measurement. Conversely, a large uncertainty does not prove that a particular observed value is far from the unknown measurand value.

If a report applies a bias correction, ask how the correction was established and how its uncertainty was carried forward. Subtracting an estimated offset does not make its estimate exact. A correction learned in one matrix or range also needs evidence before being transferred to another.

Precision can change across the analytical range

A single CV cannot describe every sample level unless that summary is justified. Constant absolute SD and constant relative SD imply different relationships between concentration and variation. Close to zero, a CV can become unstable or unhelpful because the mean appears in its denominator.

Likewise, a comparison at one reference level does not reveal whether an offset is constant or proportional. An additive difference and a scale-factor error can coincide at one point while diverging elsewhere. This is an original reasoning example: one matched point cannot uniquely determine the shape of an unknown error relationship.

Read the supported range, sample types and precision conditions before comparing headline percentages across papers. There is no universal “three replicates prove accuracy” rule, and this article does not propose one. Required evidence depends on the intended analytical claim and its consequences.

Common questions

Can a method be precise but biased?

Yes. Results can agree closely with each other while their centre is displaced from an appropriate reference value.

Is a low CV an accuracy percentage?

No. CV describes dispersion relative to the mean under stated conditions. It does not independently compare the result with a reference.

Are repeat injections equivalent to independent preparations?

No. They include different sources of variation. State which stages were independently repeated.

Does bias correction eliminate uncertainty?

No. The correction is estimated and has its own uncertainty and scope of applicability.

Does intermediate precision mean agreement between all laboratories?

No. It concerns specified changes over time within the same location and procedure. Broader reproducibility needs its own defined comparison.

See controls and replication and the peptide science library. HealthyWelness operates a commercial research-material catalog; this educational discussion makes no claim about catalog testing, certification or laboratory accreditation.

Cover: AI-generated conceptual editorial artwork, not an exact molecular structure, measured result or a photograph of HealthyWelness staff or facilities.

Sources

  1. JCGM VIM: Measurement accuracy
  2. JCGM VIM: Measurement trueness
  3. JCGM VIM: Measurement precision
  4. NIST TN 1297: Appendix D1, Terminology
  5. JCGM VIM: Intermediate precision condition
  6. NIST TN 1297: Classification of Components of Uncertainty
Chat on WhatsAppMon–Sat, 09:00–21:00 (UTC+8)