# IC50: What It Is, How to Calculate It, and IC50 vs EC50

> What IC50 means, how to calculate it with a four-parameter logistic fit, IC50 vs EC50 and Ki, a 96-well plate layout and a worked example as a CSV.

Canonical: https://origindatapartners.com/guides/ic50-calculation
By: Origin Data Partners
Status: Published
Published: 2026-11-13
Updated: 2026-11-13
Sources checked: 2026-11-13

## Quick answer

IC50 is the concentration of an inhibitor that produces half of its maximal inhibition in an assay. To calculate IC50, fit a four-parameter logistic curve to response against log concentration; the IC50 is the curve's midpoint. EC50 is the same idea for a stimulating effect. Report the fit, the controls and whether the IC50 is relative or absolute.

## Key points

- Fit a four-parameter logistic curve on a log concentration axis; do not read IC50 off a line between two points.
- Relative IC50 is the curve midpoint; absolute IC50 is where the response crosses 50% of control.
- Put vehicle and full-inhibition controls on every plate, so each plate normalizes itself.
- Keep the raw reads, the normalization and the fit settings with the result.

## What is IC50?

IC50 is the half-maximal inhibitory concentration: the concentration of a compound that produces half of its maximal inhibition of a response in a given assay. In GraphPad's plain definition, "IC50 and EC50 are, respectively, the inhibition and effective concentrations that produce 50% of maximal response" ([GraphPad FAQ 2187](https://www.graphpad.com/support/faq/how-do-i-estimate-the-ic50-and-ec50/)).

A lower IC50 means a more potent inhibitor in that assay. IC50 is not a constant of the compound: it changes with the assay conditions, such as substrate concentration, cell number, incubation time and readout, so always state them with the value. Many teams also report pIC50, the negative base-10 logarithm of the IC50 in molar units, because potencies then sit on an even scale (0.25 µM is pIC50 6.60).


## What is the difference between IC50 and EC50, and between IC50 and Ki?

EC50 is the concentration that produces half of the maximal effect of a stimulating compound, such as an agonist. IC50 is used when the compound reduces a response. The curve fitting is the same; only the direction of the curve differs.

Ki is the inhibition constant, a property of the inhibitor and its target that does not depend on the substrate concentration in the assay. For a competitive enzyme inhibitor, Cheng and Prusoff derived the relationship Ki = IC50 / (1 + [S] / Km), where [S] is the substrate concentration and Km the Michaelis constant ([Cheng and Prusoff, 1973](https://pubmed.ncbi.nlm.nih.gov/4202581/)). Use it only when its assumptions hold for your assay.

| Term | What it measures | Depends on assay conditions? |
| --- | --- | --- |
| IC50 | Concentration giving half-maximal inhibition | Yes |
| EC50 | Concentration giving half-maximal effect (stimulation) | Yes |
| pIC50 | Negative log10 of IC50 in mol/L | Yes, like IC50 |
| Ki | Inhibitor binding constant | No (in principle) |


## How do you calculate IC50? The four-parameter logistic fit

Normalize each well to the plate controls: percent activity = (signal - mean of full-inhibition controls) / (mean of vehicle controls - mean of full-inhibition controls) x 100. Then fit response (Y) against concentration (X) with the four-parameter logistic model:

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

Top and Bottom are the upper and lower plateaus, and Hill is the slope factor (positive for a falling curve written this way). Fit by nonlinear least squares, with concentration on a log scale for plotting and starting values near the data. Sebaugh's guidelines distinguish two definitions: the relative IC50 "is the parameter c in the 4-parameter logistic model and is the concentration corresponding to a response midway between the estimates of the lower and upper plateaus", while the absolute IC50 is tied to the 50% control ([Sebaugh, 2011](https://pubmed.ncbi.nlm.nih.gov/22328315/)). Say which one you report.

- Use 8 to 12 concentrations spaced evenly on a log scale (for example 3-fold steps) that span both plateaus.
- Run replicates (triplicates are common) and fit all points, not only the means.
- Fix Top or Bottom to the controls only when the data cannot define a plateau, and say so.
- Check the residuals and the confidence interval of the IC50, not only the R squared.


## How should a 96-well plate be laid out for an IC50 run?

A 96-well plate has 8 rows (A to H) and 12 columns (1 to 12). The layout in the download puts one compound in triplicate rows with an 8-point, 3-fold dilution series and both controls on the same rows, so each replicate normalizes against controls run alongside it.

| Columns | Rows A to C | Purpose |
| --- | --- | --- |
| 1 to 8 | Test compound, 10 µM down to 0.0046 µM in 3-fold steps | The dose-response series, one replicate per row |
| 9 | Vehicle only (for example DMSO at the same final percentage) | Defines 100% activity (0% inhibition) |
| 10 | Reference inhibitor at a fully inhibiting concentration | Defines 0% activity (100% inhibition) |
| 11 to 12, rows D to H | Empty in the example | Room for a second compound or a reference curve |

Edge wells evaporate faster in long incubations; many labs keep the outer ring for buffer or controls. Record the layout you used with the results.


## IC50 worked example: from plate readings to a fitted curve

The [worked example (CSV)](https://origindatapartners.com/downloads/ic50-dose-response-example.csv) is invented: a luminescence readout with vehicle wells averaging 50,257 and full-inhibition wells averaging 2,015. Each test well is normalized to percent activity with those means. The table shows the mean of the three replicates at each concentration.

- Least-squares four-parameter fit of all 24 points: Top 99.9%, Bottom 4.4%, Hill slope 1.12, relative IC50 0.25 µM (pIC50 6.60).
- At the relative IC50 the fitted response is the midpoint of the plateaus, 52.1% activity, not exactly 50%.
- The absolute IC50, where the fitted curve crosses 50% activity, is about 0.27 µM. The two differ because the bottom plateau is above zero.
- Report: relative IC50 0.25 µM, Hill 1.12, n = 3 replicates on one plate, with the plate ID and the date.

| Concentration (µM) | Replicates (% activity) | Mean (% activity) |
| --- | --- | --- |
| 10 | 7.4, 3.5, 6.2 | 5.7 |
| 3.33 | 8.1, 10.2, 10.8 | 9.7 |
| 1.11 | 21.8, 18.8, 18.0 | 19.5 |
| 0.370 | 39.3, 43.5, 42.2 | 41.7 |
| 0.123 | 70.8, 67.8, 72.0 | 70.2 |
| 0.0412 | 87.7, 90.8, 87.0 | 88.5 |
| 0.0137 | 97.8, 95.1, 95.7 | 96.2 |
| 0.0046 | 97.0, 99.6, 100.1 | 98.9 |


## Can you calculate IC50 in Excel?

Yes, with Solver. Put concentrations and responses in two columns. Put starting guesses for Top, Bottom, IC50 and Hill in four cells. In a third column, compute the predicted Y with the four-parameter formula using those cells. Compute the sum of squared differences between observed and predicted Y with SUMXMY2, then run Solver to minimize it by changing the four parameter cells, constrained to IC50 greater than zero.

Excel will not give a confidence interval this way. Dedicated curve-fitting software, R or Python report the interval and flag fits that did not converge, which is why most teams use them once a method is routine.


## What should be recorded with every IC50 so it can be checked later?

Anyone reviewing the number needs to rebuild it: the plate map, the raw reads, the control means, the normalization, the model, any fixed parameters and the software version. For studies run under GLP, a final report must include "A description of the transformations, calculations, or operations performed on the data" ([21 CFR 58.185(a)(11)](https://www.law.cornell.edu/cfr/text/21/58.185)), and raw data includes "recorded data from automated instruments" ([21 CFR 58.3](https://www.law.cornell.edu/cfr/text/21/58.3)).


## Common IC50 mistakes

- Too narrow a concentration range, so one plateau is missing and the fit invents it.
- Spacing doses linearly instead of on a log scale, which leaves too few points on the slope.
- Normalizing to controls from a different plate or day.
- Reporting an IC50 beyond the highest concentration tested as if it were measured.
- Mixing relative and absolute IC50 across a project without saying which is which.
- Keeping only the fitted value and losing the raw reads, the plate map and failed plates.


## Your dose-response archive is also a data asset

Years of plates, each with its layout, raw reads, controls, fits and the runs that failed, show how compounds actually behaved in your assays. Published papers report the winning curves; the full record is rarely public.

AI developers working on biology and chemistry ask for experiments like these with their context and their failures. Sponsor-owned studies, partner-restricted projects and identifiable human-subject data stay out unless the agreements and consents allow them; [licensing R&D lab notebooks](https://origindatapartners.com/guides/license-rd-experiment-records-ai) covers the general case.

If your company keeps years of lab notebooks and assay results, the [guide for biotechs, CROs and contract labs](https://origindatapartners.com/guides/biotech-cro-lab-data) explains what AI developers ask for and who signs. To see what yours could be worth, [describe them in six questions](https://origindatapartners.com/check-fit?sector=biotech); no files or system access needed.


## Questions and answers

### What does IC50 mean?

IC50 is the half-maximal inhibitory concentration: the concentration of a compound that produces half of its maximal inhibition of a response in a given assay. It is specific to that assay's conditions.

### What is the difference between EC50 and IC50?

EC50 is the concentration giving half of the maximal effect of a stimulating compound; IC50 is the concentration giving half of the maximal inhibition. Both come from the same kind of dose-response fit.

### How do you calculate IC50 from a dose-response curve?

Normalize each well to the vehicle and full-inhibition controls, then fit a four-parameter logistic model of response against concentration by nonlinear least squares. The fitted midpoint parameter is the relative IC50.

### Is a lower IC50 better?

A lower IC50 means a more potent inhibitor in that assay. Potency is only one property; selectivity, solubility and toxicity matter as much, and IC50 values from different assay conditions are not directly comparable.

### What is the difference between relative and absolute IC50?

The relative IC50 is the concentration at the midpoint between the fitted top and bottom plateaus. The absolute IC50 is the concentration where the response reaches 50% of the control signal. They differ when a plateau is not at 0% or 100%.

### How do you convert IC50 to pIC50?

Take the negative base-10 logarithm of the IC50 in moles per litre. An IC50 of 0.25 µM is 2.5 x 10^-7 M, so pIC50 = 6.60.

### How is Ki different from IC50?

Ki is a binding constant of the inhibitor for its target and in principle does not depend on substrate concentration; IC50 does. For a competitive inhibitor, the Cheng-Prusoff relationship Ki = IC50 / (1 + [S]/Km) links them.

### How many concentrations do you need for a reliable IC50?

Use enough log-spaced concentrations, commonly 8 to 12, to define both plateaus and the slope, with replicates at each. If a plateau is missing, extend the range where you can; fix a plateau to the controls only when you cannot, and say so.

## Sources

- [GraphPad knowledgebase: How do I estimate the IC50 and EC50? (FAQ 2187)](https://www.graphpad.com/support/faq/how-do-i-estimate-the-ic50-and-ec50/)
- [Sebaugh JL (2011), Guidelines for accurate EC50/IC50 estimation, Pharmaceutical Statistics 10(2):128-134 (PubMed)](https://pubmed.ncbi.nlm.nih.gov/22328315/)
- [Cheng Y, Prusoff WH (1973), Relationship between the inhibition constant (K1) and I50, Biochemical Pharmacology 22(23):3099-3108 (PubMed)](https://pubmed.ncbi.nlm.nih.gov/4202581/)
- [21 CFR 58.185: reporting of nonclinical laboratory study results](https://www.law.cornell.edu/cfr/text/21/58.185)
- [21 CFR 58.3: definitions (nonclinical laboratory study, sponsor, raw data)](https://www.law.cornell.edu/cfr/text/21/58.3)

## Related reading

- [Assay Data for AI: What Biotechs, CROs and Labs Can Offer](https://origindatapartners.com/guides/biotech-cro-lab-data)
- [ELN vs LIMS: Differences, What Each Stores and Exports](https://origindatapartners.com/guides/eln-vs-lims)
- [21 CFR Part 58 (GLP): Final Reports, Archives and Retention](https://origindatapartners.com/guides/21-cfr-part-58-glp)
- [Licensing R&D lab notebooks and experiment records to AI](https://origindatapartners.com/guides/license-rd-experiment-records-ai)
- [Pharmacist Side Hustle: AI Training Jobs and Pay (2026)](https://origindatapartners.com/guides/ai-training-jobs-for-pharmacists)
- [/glossary/ic50](https://origindatapartners.com/glossary/ic50)