Skip to content

How to calculate IC₅₀ / EC₅₀ online

The IC₅₀ is the concentration of a drug that produces half of its maximal inhibition; the EC₅₀ is the same idea for a stimulating (activating) drug. You get it by fitting a sigmoidal dose-response curve, usually the 4-parameter Hill equation (also called the 4PL or four-parameter logistic model), to your measurements, then reading off the curve's midpoint.

synfit does this in the browser: upload a CSV, click Fit, and read the IC₅₀ together with its 95% confidence interval.

Try it without an account

Open the demo: it starts a temporary workspace with sample datasets, deleted after 24 hours.

1. Prepare your data

You need, for each well or measurement, the concentration and the response (viability, signal, % activity…). Replicates help: they let the fit estimate the noise and give honest confidence intervals.

The simplest layout is one row per measurement:

concentration,y,replicate
0,1.00,A
0,0.98,B
0.1,0.97,A
0.1,0.95,B
1,0.71,A
1,0.74,B
10,0.17,A
10,0.19,B
100,0.04,A
100,0.03,B

A wide layout (concentrations in the first column, one column per replicate) works too. See CSV formats for both.

A few things make the IC₅₀ well determined:

  • Span the transition. Include concentrations clearly below and clearly above the IC₅₀, so both plateaus are visible. If the curve never flattens at the top dose, the bottom asymptote (and therefore the IC₅₀) is a guess.
  • Space doses logarithmically, e.g. 3-fold or 10-fold dilutions. Dose-response curves are sigmoidal on a log axis.
  • Keep the zero-drug control. It anchors the top of an inhibition curve. A log axis cannot show zero, so synfit plots the controls in a small linked panel beside the curve.
  • Raw or normalised both work. You do not have to convert to "% of control" first; the fitted top and bottom plateaus do that job.

2. Upload and fit

  1. Sign in (or open the demo) and click Upload. Choose the file; synfit detects the columns and plots the data. Add the drug name, concentration unit and direction (inhibition or activation).
  2. Check the preview plot, then Save. The dataset page shows the scatter plot and replicate heatmap, where outliers or a mis-detected column stand out.
  3. Click New analysis. The defaults suit most assays:
    • Curve model: 4-parameter Hill. Use 5-parameter only for clearly asymmetric curves (see 4PL vs 5PL).
    • Error model: Auto fits both a constant-noise and a proportional-noise model and keeps the one that explains the data better (error models).
  4. Click Fit. It takes a second or two.

3. Read the result

The analysis page shows:

Output Meaning
IC₅₀ (or EC₅₀) Concentration at half-maximal effect, in your unit, with a 95% CI
Hill Slope of the curve. 1 is the textbook value; steeper (> 1) curves switch over a narrower range
Top / Bottom The two plateaus: response with no drug, and at saturating drug
Plot Data, fitted curve and its 95% confidence band; download as PNG or SVG
Predict concentration from response Type any response (say 20 % viability) to get the concentration and its CI: an IC₂₀, IC₈₀ or any ICₓ

The IC₅₀ interval is asymmetric around the estimate (e.g. 2.2 [2.1, 2.4] µM). That is correct: potency is estimated on a log scale, so the interval is symmetric in log-concentration, not in concentration (confidence intervals).

What a wide interval tells you. If the CI spans an order of magnitude, the data do not pin the IC₅₀ down. Common causes: no plateau at one end, too few doses across the transition, or very noisy replicates. Adding doses around the midpoint usually helps more than adding replicates.

4. Outliers and re-fitting

Click a point on the plot to exclude it and re-fit. Excluded points stay visible (greyed out, with an X), so the decision is documented. You can keep several analyses of the same dataset, for example with and without an outlier, and switch between them to compare.

5. Share or reuse

  • Share link: a read-only link to one analysis, which a collaborator can open without an account (sharing).
  • Python: the fitting engine is the open-source synfit package, so the same fit can run in a script or notebook:
import pandas as pd
from synfit import SingleDrugFit, default_fit_config

df = pd.read_csv("my_assay.csv")          # columns: concentration, y, replicate
config = default_fit_config(df, direction="inhibition")
result = SingleDrugFit(df, config).fit().to_dict()
print(result["ic50"], result["ci"]["ic50"])  # estimate and 95% CI

Frequently asked

IC₅₀ or EC₅₀? Same fit, different wording. synfit labels the parameter IC₅₀ when the curve direction is inhibition and EC₅₀ for activation.

Relative or absolute IC₅₀? synfit reports the relative IC₅₀: the midpoint between the fitted top and bottom plateaus. The absolute IC₅₀ is where the response crosses 50 % of your untreated control. When the fitted bottom plateau lies below that level, enter the 50 %-of-control response in Predict concentration from response to read it off the curve. When the bottom plateau stays above it, the curve never reaches 50 % of control: the absolute IC₅₀ is not reached within the drug's maximal effect, and you should report it that way rather than extrapolate.

Can I fix the top at 100 % and the bottom at 0 %? Yes. In the analysis dialog, untick a parameter to fix it at the value you enter (fit configuration). Do this only when your controls justify it; fixing a plateau the data disagree with biases the IC₅₀.

Two drugs at once? For a combination matrix, see drug synergy analysis.