How to interpret AUC value

I am asking about the value of drug sensitivity AUC. For example, AUC values are 0.2 and -0.2. What do those mean?

I understand LFC. As Mustafa mentioned in the previous post, AUC can be considered as the average of 2^LFC. If so, then how to understand the negative AUC value? 2^LFC should be positive. Or AUC value is never negative?


As far as I am aware, AUC values should be always non-negative. Please let me know if you have an example (I believe it must be a bug then) and I am happy to take a closer look.

Thanks. I notice there are 4 different drug AUC datasets including Sanger GDSC1, GDSC2, CTD^2, and PRISM_Repurposing_Secondary_Screen. What are the difference among them?

In the repurposing screen you can get negative values because it is normalized, not an AUC measurement.
BTW I would love if this measurement were called “drug resistance” as opposed to “drug sensitivity”, since higher values correlate with greater resistance, not greater sensitivity. It had me confused for a while and I’m clearly not the only one.

@nschomer Thanks for the nice suggestion. How the values are normalized? And do you know the difference among the four datasets including GDSC1, 2, CTD^2, and PRISM_Repurposing_Secondary_Screen? And where I can find out the detailed information about them? I haven’t heard from @Mustafa_Kocak yet…


Fundamentally, the data sets are from different sources, the GDSC1, GDSC2 data sets come from the Wellcome Sanger Institute, DepMap has a brief desc. with links to the relevant papers here:

The PRISM data is from the Broad Institute, and is a particular type of screen where they mix cells together, treat them, and then count the remaining cells from each cell line using DNA barcodes, more info at

I’m not sure exactly what the basis for normalization is, but the axis legend calls it a “log2” fold change, so I’m assuming a -1 = twice the drug sensitivity of a 0, and 4 times the sensitivity of a 1, etc.
Hope this helps.

1 Like

For the PRISM data AUC values should be positive too, the log2 fold change data. is presented per dose (2.5 uM for PRISM Primary and 8 different dose points for PRISM secondary).

GDSC is the dataset generated by Sanger institute, while CTD2 and PRISM are generated at Broad Institute (through different projects and assays) and as nschomer mentioned the relevant papers are listed in the download page.

@nschomer @Mustafa_Kocak Thank you so much for your help! I think that I am more clear now.