In this update we have changed the way numbers with modifiers (that is: >, <, ≥ and ≤) are aggregated. Numbers with modifiers are also referred to as bounded values, and this will mainly affect averages of IC50 measurements. The rule can be stated as: a bounded value will not be included when computing the average if there is another measurement that is consistent with it and is more specific. Exact values are always included.
- Less specific bounded values are not included:
Average(<1, <10) = <1
Average(20, 30, <100) = 25
Average(80, <100, 150) = 115 (<100 is consistent with 80 and less specific)
- Bounded values are retained if they are not consistent with an exact value:
Average(<2, 8) = <5
This rule is based on how drug discovery data is normally generated. For example, if you perform a concentration-response assay between 100 to 10,000 nM, and the IC50 comes out as <100 nM, you will test at lower concentrations, say between 1 and 200 nM. If the IC50 in the second test was 80 nM, it makes sense to leave out the <100 nM from the first assay, because it has been superseded by the more specific 80 nM measurement.
On the other hand, if you ran the 100 to 10,000 nM assay twice and the only IC50s you had were <100 and 120 nM, we shouldn’t ignore the <100 nM, because doing so would result in a number that was artificially high.
All existing data has been automatically recalculated following the new rule, which applies not only to averages, but to all aggregations, such as median and geometric mean.