gms | German Medical Science

In many toxicological assays or when measuring gene expression data, a response variable is measured for increasing concentrations of a compound and a negative control.

A dose-response-curve (DRC) can be fitted to determine the concentration where a specific effect level is attained.

For viability assays, data are usually normalized so that the response value of the controls corresponds to 100%.

In practice, a deviation of the controls in comparison to the response values of other concentrations in the no-toxicity range can often be observed.

This leads to an upper asymptote that does not correspond to 100% and therefore to an inability to properly calculate and interpret concentrations where a specific effect is obtained.

In a simulation study we analyse four different methods for dealing with the problem of deviating control values, including re-normalisation, forcing the upper asymptote through 100%, omission of the controls and a possibility to model negative deviations.

A decision rule is derived which of the presented methods should be used, depending on the choice of concentrations measured, the variance of the replicates per concentration, and the deviation of the controls.

In a broader context, the concentration is searched where the effect exceeds a pre-defined threshold.

This can be done in a discrete way by only considering the concentrations at which the effect was actually measured or in a continuous way by first fitting a DRC.

When not only the concentration at which a threshold is exceeded is of interest, but the concentration at which a threshold is exceeded significantly, the t-test or the Dunnett-procedure that adjusts for multiplicity can be used for discrete testing of the concentrations.

We propose a normal-distribution based test that can be applied to a fitted DRC to assess whether the response at a specific concentration significantly exceeds a threshold compared to the response for the negative control.

In a simulation study, we compare the classical t-test, the Dunnett-procedure and the newly proposed test in different situations of real concentration-response-relations.

Both simulation studies are based on the assumption that the true relation between concentration and response is sigmoidal-shaped and can therefore be modelled by a four-parameter log-logistic function.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.