I have tried to use the drug sensitivity datasets for GDSC1 and GDSC2 both for IC50 and AUC. I have noticed that the AUC dataset is richer than the IC50 one (as expected since AUCs can be calculated using less datapoints than those needed to calculate a proper IC50), so I’d rather be interested in using the AUC one. Just to confirm that AUC and IC50 could be used indistinctly, I checked the correlation between AUC and IC50 (actually pIC50) values for a few compounds and I found out that, within the plot, there were three correlation groups like three separated “datapoints clouds” within the same range of AUC vales. To me, this suggests that three different scales were used in the plots for the dose response curves depending on the IC50 vaue, i.e., three different range of concentrations (starting from 10 mM, 100 uM or 1 uM as the highest concentrations) and this has not been corrected in the dataset.
If this is the case the AUC values cannot be used as a surrogate of potencies since the same AUC may correspond to IC50s spanning up to 6 orders of magnitude. For instance, an AUC of 0.81 is found for vimblastine in cell line ACH-000004 (IC50 14 nM) and for AICAR in cell line ACH-000258 (IC50 23 mM). Is there a way this can be corrected? In other words, is it possible to obtain the concentration range each compound has been tested?
