Reading ChEMBL: comparing measurements from drug research

Comparing drug research measurements in ChEMBL: Nadoo Bio public database series

ChEMBL is a public database of experimental results describing how compounds act on proteins, cells, and other systems, together with information from drug research. It records what was tested, how it was tested, and the outcome, with links to the original sources. It can support comparisons between drug candidates and preparation of training data for predictive models. [1]

If two candidates have experimental values of 10 and 100, which is better? The numbers alone cannot tell us. Interpretation depends on what was measured, whether the units match, and the test conditions. ChEMBL lets you find this context alongside the measurements.

Four pieces of information behind a measurement

Here, activity means an action or response observed in a defined test. Different tests observe different things, such as binding to a protein or reducing a particular response.

Read ChEMBL by connecting the compound, assay, target, and measurement. The compound tells you what was tested; the assay describes how it was measured; the measurement gives the result. The target, or object of the test, can range from a specific protein to a more complex system such as a cell. [2]

Connections between a compound, assay, target, measurement, and source

Knowing what was measured and how makes the number interpretable.

One compound can have multiple values from different assays. Keeping the source and method attached helps explain where each value came from. A table containing only compound names and numbers loses this context.

IC50 measures inhibition; Kd describes binding strength

IC50 is the concentration needed to reduce the response in a defined assay by half. Under the same assay conditions, a smaller value means the response was inhibited at a lower concentration. IC50 depends on the materials and conditions used, so values from different experiments may not be directly comparable.

Kd, the dissociation constant, describes binding strength when association and dissociation between two molecules are at equilibrium. Under comparable conditions, a smaller value indicates stronger binding.

The two measures answer different questions. Kd describes how tightly molecules bind; IC50 describes how low a concentration was needed to reduce a specified response. A molecule can bind well without changing the desired response, and cellular responses involve multiple processes. IC50 and Kd therefore should not be mixed as though they were the same kind of measurement.

A smaller number can represent a higher concentration

Concentration describes how much of a substance is present in a given volume. Nanomolar (nM) and micromolar (µM) concentrations are related by 1 µM = 1,000 nM. The figure below uses hypothetical values to explain the units.

Converting hypothetical values of 10 nM and 1 µM to 10 nM and 1,000 nM

Hypothetical example: expressed in the same unit, the values are 10 nM and 1,000 nM.

Looking only at the numbers, 10 is greater than 1. But 1 µM equals 1,000 nM, so its concentration is much higher than 10 nM. To compare which candidate acted more effectively, first align units, then compare within the same kind of assay and compatible conditions.

Symbols attached to measurements also carry information. >10 gives a lower bound; it does not say that 10 was measured. Replacing every such value with 10 changes the meaning of the record. This is why those symbols should also be preserved in model-training tables.

Even records under the same protein name may differ in species, the region used in the experiment, or mutations. Tests using isolated proteins and tests using cells answer different questions. Collecting this context helps define which values can be compared.

Reading small concentrations with pChEMBL

Very small concentrations are hard to scan when units and numbers of digits differ. ChEMBL's pChEMBL converts certain eligible measurements to a negative logarithmic scale based on molar concentration. The scale makes smaller concentrations appear as larger numbers: a higher pChEMBL value corresponds to a lower original concentration. [2]

Changing the scale does not change what the original assay measured. A transformed IC50 still describes inhibition, and a transformed Kd still describes binding. Not every measurement receives a pChEMBL value.

Duplicate records from the same source, data-validity flags, and the assigned target also affect the comparison. [2] Restricting the target, species, and assay family consistently makes it easier to collect comparable data. Recording inclusion and exclusion criteria allows later analyses to follow the same choices.

What makes experimental values comparable

In ChEMBL, the conditions behind a value matter as much as its magnitude. Preserving the links between compound, assay, target, and source helps select records that can be compared on a common basis. For predictive models, this information also defines what experimental outcomes are used for training and evaluation.

Selectivity—the extent to which a compound acts preferentially on the intended target—toxicity, and effects in humans each require appropriate experimental evidence. Understanding the question one measurement answers also clarifies what evidence to seek next.

Terms of use

ChEMBL data may be used for both research and commercial purposes. Under its CC BY-SA 3.0 license, BY requires attribution, and SA requires adaptations that are shared to carry the same license terms. Provide attribution and a license link; when distributing adapted material, identify the changes and apply the same terms. [1] [3]

Some molecular properties calculated using commercial software have additional conditions. The official FAQ states that those calculated values must not be extracted for training a model that imitates the original commercial calculation process. [1]

References

  1. ChEMBL. General Questions. Official FAQ. Checked October 23, 2026.

  2. ChEMBL. Assay and Activity Questions. Official FAQ. Checked October 23, 2026.

  3. Creative Commons. Attribution-ShareAlike 3.0 Unported (CC BY-SA 3.0). License deed. Checked October 23, 2026.