Is there a threshold in Drug Sensitivity AUC?

Q&A
1

捕获
捕获1259×615 31.9 KB

Is there a threshold in Drug Sensitivity AUC, below which it is considered that the drug has an impact on cell line proliferation? which threshold?
Similarly, does a threshold also exist in the PRISM Repurposing Primary Screen dataset?

2

Hi,

For the Primary screen, the values are log2(Viabilities) at 2.5uM, and historically, we considered 30% viability (corresponding to log2(0.3) =~ -1.74 as a “sensitivity threshold.”

For AUC, it is trickier since it is also a function of the dose range screened for the drug. I don’t know the particular analysis you are following, but my recommendation would be scaling AUCs (maybe z-scores) and treating cell lines on the left tail (maybe z < -2) as sensitive lines.

3

Are we certain this is correct? The DepMap paper states “We label cell lines as sensitive to a treatment if the median-collapsed fold-change is less than 0.3.”

Thus, it appears that the binary cutoff is actually log2(0.3), which is 1.23. Meaning that a 23% increase in cell line proliferation relative to the control. I am unclear why an increase relative to the control would be considered sensitive to the treatment, although I think I read about noise being an issue.

If the metric were indicating 30% viability, the wording should state this, and not specifically mention “median-collapsed fold-change is less than 0.3.”

Can I get a clarification from the Depmap team on this important point?

Thanks

4

Apologies, I of course meant that since a “median-collapsed log-fold change” of 0.3 is the value reported, which itself is log2(1.23). So according to the paper, the reported value of 0.3 used as a threshold would correspond to cell proliferation relative to DSMO of 1.23, ie. a 23% increase.

Can someone from the DepMap team please clarify this?

5

Actually log2(0.3) is around -1.74, the fold change is calculated as treatment_MFI/negative_ctrl_MFI, so 0.3 means it is only 0.3 of negative, indicating decreased viability.

6

Thanks, but the paper clearly states that the threshold is a “median-collapsed log-fold change” of 0.3, ie. after log2 is applied.

I get your point, that a fold-change (treatment_MFI/negative_ctrl_MFI) of 0.3 gives a log2 value of -1.74. But that is not what the paper states.

So, do we assume that the paper used inaccurate language, and that the threshold was actually a fold-change (treatment_MFI/negative_ctrl_MFI) of 0.3 before log2 transformation, and hence a “median-collapsed log-fold change” of -1.74?

7

In their paper (https://doi.org/10.1038/s43018-019-0018-61) or PMID: 32613204, they explicitly state in the Method section, under Data processing, the last sentence: “We labeled cell lines as sensitive to a treatment if the median-collapsed fold change was < 0.3”, so I believe it is not on the log scale

8

Hi, my apologies. I had it quoted as “median-collapsed log-fold change”.

Now I looked again and saw that I had taken that from elsewhere in the paper.

I stand corrected!