Parallel-line analysis focuses on a linear response region
A classical parallel-line assay relates response to transformed dose over a range where the relationship is adequately linear. Reference and test lines share a slope under the parallel model, and horizontal displacement supports relative potency.
Linearity and parallelism need evaluation under the selected design and statistical framework. Restricting analysis to a linear region can be appropriate but should be predefined rather than selected post hoc for each run.
4PL and 5PL models use the broader sigmoid
A 4PL models lower and upper asymptotes, midpoint, and slope; a 5PL adds asymmetry. Relative potency may be estimated through midpoint relationships or a joint/restricted nonlinear model.
Nonlinear models can use more of the response range, but poorly supported asymptotes or asymmetry can create unstable parameters. Shared-parameter assumptions still need scientific and diagnostic support.
Compare the analytical questions, not model popularity
Consider response shape, informative range, number and placement of doses, replicates, variance, ability to meet assumptions, historical performance, interpretability, and validation evidence.
| Dimension | Parallel-line | 4PL/5PL |
|---|---|---|
| Response region | Selected approximately linear region | Sigmoidal range where parameters are supported |
| Core comparison | Horizontal displacement with common slope | Midpoint or joint nonlinear curve relationship |
| Key diagnostics | Linearity, parallelism, residuals | Coverage, residuals, parameters, shared-shape assumptions |
| Primary risk | Post hoc range selection | Overparameterization or unsupported asymptotes |
Report assumptions and restrictions explicitly
For either approach, include transformed variables, selected doses, model equation, variance and weighting, restrictions, estimates and uncertainty, diagnostics, suitability criteria, exclusions, units, reference assignment, and software and settings versions.
For controlled routine use, each model should have a named, versioned calculation with defined input requirements, outputs, diagnostics, and expected-result tests that cover normal and edge cases.
Frequently asked questions
Is 4PL always more modern and therefore better?
No. Model choice depends on the analytical response, design, assumptions, intended result, and evidence—not chronology.
Does parallelism mean the plotted curves look parallel?
Visual inspection is useful but not sufficient. Define the statistical or model-based comparability assessment and its acceptance rule.
Can both models be available in one product?
Yes, if each calculation and its method settings are versioned and tested. Routine methods should control which model is used.