White paper

Can these IC50 values actually be compared?

Same target. Same IC50 label. Same units. Still not the same experiment.

In a 2024 analysis of ChEMBL32, 64% of paired IC50 values for the same compounds differed by more than 0.3 log units (about twofold), and 27% differed by more than tenfold. Those values were not necessarily wrong. The assays were not the same.

  • Why IC50 is assay-dependent. The measured response and the experimental conditions are part of the value.
  • Where context disappears. Standardization of names and units does not make two assays equivalent.
  • What a usable record keeps. Target, system, conditions, method, and readout stay attached to the number.