Skip to main content
Delta Life Science

KD vs IC50: how binding affinity and inhibition differ

Line chart showing the IC50 of a competitive inhibitor with Ki 10 nM rising from 10 nM to 110 nM as substrate rises to ten times Km
In short

KD describes how tightly two molecules bind at equilibrium. IC50 describes how much inhibitor halves a signal in one specific assay, so it moves when the assay changes. Convert IC50 only when the mechanism and conditions allow it, and measure KD directly when you need to compare compounds across assays.

KD vs IC50: two numbers for two different questions

Drug discovery teams describe a compound with two kinds of numbers: how tightly it binds its target, and how well it blocks a function. The KD vs IC50 question comes up when someone puts the two side by side and they do not match.

The equilibrium dissociation constant, KD, belongs to a binding pair. It is the concentration of free molecule at which half of its binding partner is occupied at equilibrium, given in molar units. A KD of 10 nM means half of the target is bound when 10 nM of the compound is free in solution. Lower KD means tighter binding.

Simulated steady-state binding curve with half of the binding sites occupied at a KD of 10 nM
Figure 1. Illustrative 1:1 binding curve. Occupancy reaches 50% at the KD and about 90% at nine times the KD.

KD also connects to speed. The association rate constant, kon (1/M/s), says how fast the complex forms. The dissociation rate constant, koff (1/s), says how fast it falls apart. KD equals koff divided by kon.

The half maximal inhibitory concentration, IC50, belongs to an assay. It is the inhibitor concentration that reduces a measured signal, such as enzyme activity or the binding of a partner protein, to half. Change the assay and the same compound can give a different IC50 [2].

ValueWhat it describesChanges with assay conditions?How you get it
KDBinding of two molecules at equilibriumLittleDirect binding: titration or kinetics
KiThe KD of an inhibitor for its targetLittle, if the assumed mechanism is rightCalculated from IC50, or fitted from inhibition data
IC50Inhibitor concentration that halves a signalYes: substrate or tracer, target concentration, timeDose-response curve in one assay

Even KD is not a pure constant. Buffer composition and temperature change measured affinities, so report both with every value [3].

IC50 moves when the assay changes

Most KD vs IC50 mismatches start with the substrate. A competitive inhibitor fights the substrate, or a labeled tracer, for the same site. Add more substrate and you need more inhibitor to reach half inhibition. The Cheng-Prusoff equation describes this: IC50 = Ki × (1 + [S]/Km), where [S] is the substrate concentration and Km is the substrate's Michaelis constant, the substrate concentration that gives half the maximal reaction rate [2].

Line chart showing the IC50 of a competitive inhibitor with Ki 10 nM rising from 10 nM to 110 nM as substrate rises to ten times Km
Figure 2. One inhibitor, one Ki, six IC50 values. At [S] = Km the IC50 is exactly twice the Ki. Source: Kalliokoski et al., 2013 [2] (calculated with the Cheng-Prusoff equation).

Substrate is not the only lever:

  • Target concentration. A high protein concentration raises the apparent IC50 (see the depletion check below).
  • Signal window. A competition assay needs enough tracer signal, and that can set an upper limit on the affinity it can report [3].
  • Incubation time. A competition read before equilibrium gives a value that depends on when you read it [4].
  • Mechanism. Allosteric and other noncompetitive inhibitors do not shift with substrate the way competitive ones do [2,3].
  • Cells. A cell-based IC50 adds uptake, protein binding and downstream signaling on top of binding.

How do you convert IC50 to Ki or KD?

For a competitive inhibitor measured at equilibrium, divide the IC50 by (1 + [S]/Km), or by (1 + [L]/KD) for a labeled tracer at concentration [L]. The result is Ki, the inhibitor's own dissociation constant for the target. Outside these conditions the equation still returns a number, but not a reliable one [2,4].

Run the substrate at its Km and the correction is a factor of 2: Ki is half the IC50. Many screening assays use this balanced condition because it detects inhibitors with different mechanisms [2].

The equation also assumes that both tracer and inhibitor are in large excess over the protein. When protein is not well below the inhibitor, it binds up inhibitor and the apparent IC50 rises. In a simulation study, the simple equation lost accuracy when this condition failed, while an exact equation (Wang's) gave the best Ki estimates across all conditions tested [4]. Use a Km or tracer KD measured in the same buffer and at the same temperature as the assay. Run these checks before you convert, and test whether the inhibitor is competitive or allosteric first [3]:

Flow of five checks before converting IC50 to Ki: mechanism, equilibrium, depletion, matched constants and direct confirmation
Figure 3. Depletion matters when the protein concentration approaches the Ki: the simple equation then overestimates it.

Can you compare IC50 values from different studies?

Only roughly. In an analysis of public IC50 data for identical protein and compound pairs, about 68% of independent measurements agreed within a factor of 4.8 [2]. That is good enough for large data sets, but too coarse to rank two close analogs.

Within one lab and one assay, the spread is much smaller. The same study found standard deviations of 0.22 and 0.17 log units for two reference inhibitors measured repeatedly in-house, against 0.68 log units for mixed public data [2].

Bar chart comparing the spread of IC50 values: 0.68 log units for mixed public data against 0.22 and 0.17 for repeated in-house measurements
Figure 4. A standard deviation of 0.68 log units means about one in three IC50 values for the same pair differs by more than a factor of 4.8. Source: Kalliokoski et al., 2013 [2] (ChEMBL pairs versus repeated in-house reference measurements).

The authors also traced many large differences to database errors: wrong units, wrong receptor subtypes and cell assays filed as biochemical ones. Across the database, Ki values were on average about 2.3 times lower than IC50 values for the same pair, close to the factor of 2 expected at [S] = Km [2].

For direct affinity measurement, see small molecule binding affinity.

Kinetics show what KD and IC50 both hide

Two compounds with the same KD can behave very differently. In the simulation below, kon is 1 × 10⁵ 1/M/s and koff is 0.001 1/s, which gives a KD of 10 nM. A compound with kon 1 × 10⁶ 1/M/s and koff 0.01 1/s has the same KD, but leaves ten times faster.

Residence time, 1/koff, is how long a complex lasts on average [5]. At koff 0.001 1/s it is about 17 minutes. At 0.01 1/s it is under 2 minutes. Slow dissociation can keep a target occupied after free drug levels fall, which no equilibrium number shows [6].

A label-free biosensor measures these rates directly. One binding partner, the ligand, is fixed to a sensor surface. The other, the analyte, flows over it, and the instrument records binding in real time as a sensorgram: a rising association phase while analyte flows, then a falling dissociation phase in buffer. Fitting a concentration series gives kon and koff, and from them KD [1].

Simulated sensorgram for five analyte concentrations with slow association and slow dissociation, KD 10 nM
Figure 5. Illustrative 1:1 sensorgram, kon 1 × 10⁵ 1/M/s and koff 0.001 1/s. After 15 minutes in buffer, about 40% of the complex is still bound.

Methods in this class include surface plasmon resonance (SPR), biolayer interferometry (BLI), grating-coupled interferometry (GCI) and Nanophotonic Evanescent field Sensing (NES), the integrated photonics method behind Delta's inQuiQ.

Measure both numbers on one biosensor

The same instrument can give you KD and IC50. For KD, fix the target and inject the compound at several concentrations. For IC50, premix the target with a dilution series of inhibitor, then inject the mixtures over a surface that carries the target's binding partner. Free target binds the surface and target held by the inhibitor does not, so the signal drops as inhibitor concentration rises [7].

PD-1 and PD-L1 make the KD vs IC50 difference concrete. Delta application notes report a KD of 37.7 nM for the PD-1/PD-L1 interaction [8] and an IC50 of 8.2 nM for its inhibition by BMS-1166 [9]. The KD describes the protein pair. The IC50 describes the small molecule in that assay, and it is not the KD of BMS-1166 for its target.

Choose the number by the question you ask:

  • IC50 to rank compounds inside one assay series
  • Ki or KD to compare across assays, labs or targets
  • kon and koff when how long a compound stays bound matters

Questions researchers ask

What is the difference between IC50 and EC50?

EC50 is the concentration that gives half of the maximal effect, whether the compound activates or inhibits. IC50 is used for inhibition only. For a compound that inhibits fully, the two are the same. For a partial inhibitor that never reaches 50% inhibition, no IC50 exists, but an EC50 can still be fitted.

Is a lower IC50 always better?

Within one assay, a lower IC50 means a more potent compound. It says nothing about selectivity, how long the compound stays bound or how it behaves in cells. A slightly weaker compound with a slower off-rate or a cleaner selectivity profile can be the better lead, so judge potency together with those properties.

How many concentrations do you need for a reliable IC50?

Use 8 to 12 concentrations, spaced two to three fold, so the curve shows a clear top and bottom plateau. Center the series on the expected IC50. Without both plateaus the fit has to guess the maximum and minimum signal, and the IC50 can shift by a large factor between repeats.

What does the Hill slope of an IC50 curve tell you?

Simple 1:1 inhibition gives a Hill slope close to 1, with 10% to 90% inhibition over about two log units. A much steeper curve can point to aggregation or tight binding, a much shallower one to several binding sites. To test it, refit with the Hill slope fixed at 1 and compare the two IC50 values.

References

  1. Real-Time Analysis of Biomolecular Interactions. Springer Japan. 2000. doi:10.1007/978-4-431-66970-8
  2. Kalliokoski T, Kramer C, Vulpetti A et al.. Comparability of mixed IC₅₀ data - a statistical analysis. PLoS One. 2013. doi:10.1371/journal.pone.0061007
  3. Hulme EC, Trevethick MA. Ligand binding assays at equilibrium: validation and interpretation. Br J Pharmacol. 2010. doi:10.1111/j.1476-5381.2009.00604.x
  4. Du Y. Binding Curve Viewer: Visualizing the Equilibrium and Kinetics of Protein-Ligand Binding and Competitive Binding. J Chem Inf Model. 2024. doi:10.1021/acs.jcim.4c00130
  5. Bernetti M, Masetti M, Rocchia W et al.. Kinetics of Drug Binding and Residence Time. Annu Rev Phys Chem. 2019. doi:10.1146/annurev-physchem-042018-052340
  6. Tonge PJ. Drug-Target Kinetics in Drug Discovery. ACS Chem Neurosci. 2018. doi:10.1021/acschemneuro.7b00185
  7. Karlsson R, Karlsson O, Belcher P. Fragment and Low Molecular Weight Compound Analysis. Handbook of Surface Plasmon Resonance. 2017. doi:10.1039/9781788010283-00255
  8. Delta Life Science. PD-1/PD-L1 Interaction. Application note. Link
  9. Delta Life Science. PD-1/PD-L1 Inhibition with BMS-1166. Application note. Link
Published 6 November 2026 · Updated 6 November 2026More in Label-free 101

Cookie settings

Choose which optional tools we may use. You can change your choice at any time through "Cookie settings" at the bottom of every page. Privacy policy

Strictly necessary

Always on

Needed for the website to work, such as remembering your cookie choice. These cannot be switched off.

  • Cookie consent choice: Stores your consent choice in your browser so we do not ask again on every page.

Privacy-friendly statistics

Always on

Anonymous visitor statistics with little or no privacy impact. IP addresses are anonymised, no cross-site profiles are built, and the data is not shared for other purposes.

  • Matomo: Visitor statistics on our own server, with anonymised IP addresses and no data shared with others. Works without cookies; uses a cookie to recognise returning visitors only if you accept Analytics, or where local law allows this for privacy-friendly statistics.
  • Ahrefs Web Analytics: Cookie-free visitor statistics. No personal data is collected.

Analytics

Helps us understand how visitors use the site, for example through heatmaps or session recordings. These tools may set cookies and process your IP address.

  • Microsoft Clarity: Heatmaps and session recordings that show how visitors navigate the site. Provider: Microsoft.

Marketing

Used to measure and improve our advertising, for example on Google or LinkedIn. These tools may set cookies and share data with the advertising platform.

  • Google Ads conversion tracking: Measures which Google ads lead to enquiries. Provider: Google.
  • LinkedIn Insight Tag: Measures the results of LinkedIn campaigns. Provider: LinkedIn.
  • OpenAI Ads Pixel: Measures the results of ChatGPT advertising campaigns. Provider: OpenAI.

Region detection: IP Geolocation by DB-IP