Most unreliable IC50 values cannot be fixed by better curve fitting. They come from the experimental design. If the concentrations do not cover both plateaus, or the points are spaced too far apart around the midpoint, no model can produce a reliable estimate. This guide covers the design decisions you make before you pipette anything.

1. The goal: define both plateaus and the transition

A four-parameter logistic (4PL) curve has four parameters:

Y = Bottom + (Top - Bottom) / (1 + (conc / IC50)^Hill)

To estimate all four, your data must show:

a top plateau: concentrations with no measurable effect,
a bottom plateau: concentrations with maximal effect,
at least 3 points on the steep part of the curve, between roughly 20% and 80% effect.

If either plateau is missing, the fit has to extrapolate it, and the IC50 then depends on that guess. That is the most common reason for very wide confidence intervals.

2. Start with a range-finding experiment

When you do not know the potency of a compound, do not go straight to a full replicate design.

Run a wide screen: for example 8 points at 10-fold spacing (e.g. 100 µM down to 10 pM).
Find the region where the response changes.
Design the definitive experiment around that region with tighter spacing.

A range finder costs one plate and saves several plates of uninterpretable data.

3. Choose the dilution factor

The dilution factor sets how densely you sample the transition.

Dilution factor	Span of 8 points	Best for
10-fold	7 log units	Range finding only
~3.16-fold (half-log)	3.5 log units	Unknown or broad curves
3-fold	~3.3 log units	Standard definitive assay
2-fold	~2.1 log units	Steep curves, known IC50

Rule of thumb: the steeper the curve (Hill slope above about 1.5), the smaller the dilution factor you need. Otherwise the whole transition can fall between two neighbouring concentrations.

Log spacing is not optional. Linear spacing (10, 20, 30, 40 µM...) puts most points on one plateau and almost none around the IC50.

4. How many concentrations?

8 to 12 concentrations is the practical standard for a definitive IC50.
Fewer than 6 points leaves too few degrees of freedom for a 4PL fit with any confidence.
Centre the series on the expected IC50 so that roughly half the points lie on each side.

5. Replicates: technical vs biological

Technical replicates (duplicate or triplicate wells) catch pipetting errors and allow outlier detection.
Biological replicates (independent experiments on different days or cell passages) tell you how reproducible the IC50 really is.

Report the IC50 from at least 3 independent experiments. Average them on the log scale (mean of log IC50, or pIC50), not the linear scale, because IC50 values are log-normally distributed.

6. Controls you cannot skip

Vehicle control (0% effect): cells or enzyme with solvent only. This defines the top.
Maximal-effect control (100% effect): a reference inhibitor at a saturating dose, or no-cell/no-enzyme wells. This defines the bottom.
Reference compound: a known inhibitor on every plate to check plate-to-plate consistency.
Constant solvent concentration: keep DMSO the same in every well (typically 0.1–0.5%). A serial dilution in buffer that also dilutes DMSO introduces a second variable.

Normalizing to these controls lets you constrain the top and bottom when a plateau is poorly defined. Only do this when the controls really represent the same 0% and 100% as the curve.

7. Plate layout

Do not put the whole dilution series in one edge row. Edge wells evaporate faster and bias one end of the curve.
Fill outer wells with buffer, or randomize the position of replicates.
Spread controls across the plate, not only in column 1.

8. Check the design before fitting

Before you trust a number, check:

Are both plateaus defined by at least two points each?
Is the fitted IC50 inside the tested concentration range?
Is the 95% confidence interval of the IC50 narrower than about one log unit?
Is the Hill slope plausible for your mechanism (often around 0.5 to 2)?

If the IC50 lies outside the tested range, report it as greater than or less than the top or bottom concentration. Do not report it as a value.

Common design mistakes

Using a 10-fold series for the final experiment.
Stopping the series before the bottom plateau because the compound is insoluble. Note the solubility limit instead.
Changing DMSO concentration across the series.
Pooling IC50 values on a linear scale.
Reporting an extrapolated IC50 without flagging it.

Doing it in Labreador

The Dose-Response module fits 4PL and 5PL models, reports the IC50/EC50 with a 95% confidence interval, and warns you when an asymptote is extrapolated or the IC50 falls outside the tested range. If you need to compare a test sample with a reference, Parallel Line Analysis checks parallelism and calculates relative potency. Everything runs in your browser, so no data leaves your machine.

If you use Labreador in a publication, please cite:

Helczman, M. (2026). Labreador - Bioassay Analysis Platform (Version 1.6.0) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.21676846

How to Choose Concentrations for an IC50 Dose-Response Experiment