Skip to content

4PL vs 5PL: choosing a dose-response model

The 4-parameter logistic (4PL) model, the Hill equation, is the default for dose-response curves: a top plateau, a bottom plateau, a midpoint (IC₅₀ or EC₅₀) and a slope (the Hill coefficient). Its S-shape is symmetric on a log-concentration axis: the curve leaves the top plateau exactly as it approaches the bottom one.

The 5-parameter logistic (5PL) model adds an asymmetry parameter, so one shoulder of the curve can be sharper than the other. synfit offers both: choose Curve model → 4-parameter / 5-parameter Hill in the analysis dialog.

When to use which

Start with 4PL. It is the standard in pharmacology, its IC₅₀ is directly the half-maximal concentration, and its four parameters are usually all well determined by a normal 8–12 point dilution series.

Consider 5PL when:

  • the residuals plot shows a systematic pattern: points sit above the 4PL curve on one shoulder and below it on the other, consistently across replicates;
  • the assay is known to give asymmetric curves (common for immunoassays such as ELISA standard curves);
  • you have many concentrations densely covering both shoulders. The asymmetry parameter is only identifiable when both bends of the curve are well sampled.

Stay with 4PL when the data are sparse or noisy. With few points, 5PL fits the noise: the curve looks a little closer to the points, but the parameter confidence intervals become wide and the potency estimate less stable.

The 5PL potency is not the IC₅₀

In the 5PL model the concentration parameter is no longer the midpoint of the curve, so synfit labels it κ (kappa), not IC₅₀. The two are related by

\[ C_{50} = \kappa \cdot \left(2^{1/S} - 1\right)^{1/n_H} \]

where \(S\) is the asymmetry and \(n_H\) the Hill slope. With \(S = 2\) and \(n_H = 1\), the half-maximal concentration is about 0.41 κ, a factor of 2.4.

To get the IC₅₀ from a 5PL fit, use Predict concentration from response on the analysis page and enter the response halfway between the fitted top and bottom. The table returns that concentration (inverse prediction). Treat its interval as approximate: it is read off the curve's band at a fixed response, not propagated from the uncertainty of κ, the slope and the asymmetry.

Do not compare a 5PL κ with a 4PL IC₅₀ from another experiment or paper; compare half-maximal concentrations.

Comparing the two on your data

synfit keeps any number of analyses per dataset, so you can fit both and switch between them:

  1. Fit with 4-parameter Hill.
  2. Click New analysis again, choose 5-parameter Hill, and fit.
  3. Compare the residuals plots. If the 4PL pattern disappears and the parameter CIs stay reasonable, that supports using 5PL. If the residuals look the same, keep 4PL.
  4. Compare the half-maximal concentration from both (via inverse prediction for 5PL). If the residuals did not justify 5PL, report the simpler 4PL result.

In drug combinations

For a combination matrix you can choose 4PL or 5PL per drug. The synergy reference surfaces then use each drug's own curve shape, asymmetry included. Use 5PL there only when that drug's single-agent edge clearly needs it: the edges usually have fewer points than a dedicated single-drug experiment.

In short

4PL 5PL
Parameters top, bottom, IC₅₀, Hill + asymmetry; potency reported as κ
Curve shape symmetric on log axis one shoulder sharper
Data needed 6–12 well-spread doses dense coverage of both shoulders
Potency IC₅₀ read directly IC₅₀ via inverse prediction
Default yes only when residuals demand it

Further reading