search

bhklab / PharmacoGx

R package to analyze large-scale pharmacogenomic datasets.
http://pharmacodb.pmgenomics.ca
GNU General Public License v3.0
65 stars 26 forks source link

ic50 published values & Transformation to get recomputed values? #52

Open e-arslan opened 5 years ago

e-arslan commented 5 years ago

I couldn't find ic50_published for ctrpv2. How can I get those values?

CTRPv2ic50_p <- summarizeSensitivityProfiles(CTRPv2, sensitivity.measure='ic50_published')
Error in summarizeSensitivityProfiles(CTRPv2, sensitivity.measure = "ic50_published") : 
Invalid sensitivity measure for CTRPv2, choose among: auc_published, ec50_published, auc_recomputed, ic50_recomputed

IC50 values are between 0 and 8 in other studies but ic50_recomputed has a huge range. An example of a cell line:

     Min.   1st Qu.    Median      Mean   3rd Qu.      Max.      NA's 
  0.01711  12.57788 166.31685       Inf       Inf       Inf       122

A good threshold for ic50 is called 1 or 1.2 microM, what might be a good threshold for this range?

What is the transformation you applied to get new ic50 values or AUC recomputed values from originally published values?

p-smirnov commented 5 years ago

Hi @e-arslan

The range between 0-8 is usually due to the fact that IC50 values are artificially truncated at the maximum concentration tested.

We do not apply a direct transformation to the published values. Rather, we refit the HIll Slope curves and report the actual IC50 from the curve fit, when it is defined (the bottom asymptote of the curve must be below 50% inhibition. If it is at 50, mathematically the IC50 works out to be positive infinity, as you see in some cases in CTRPv2). However, estimation of the IC50 extrapolated above the maximum concentration tested is often unstable (as small deviations in a single point can have a large effect of the value). My recommendation therefore would be to treat the IC50s above the maximum tested concentration either as NAs, or as right censored values (akin to survival analysis).

For published CTRPv2 values, I don't believe the publication used IC50 values, but I will double check. We do however include the published EC50 values. The difference between the two is: EC50 is concentration where 50% of maximum inhibition is reached, while IC50 is where 50% inhibition is reached. EC50 is defined unless the curve is flat, while IC50 is defined only if maximum effect is more that 50% inhibition.

I cannot recommend any arbitrary threshold for a good IC50, this would depend complexly on the drug, model system, and other details about the disease or biological process you are studying. I would comment however that the IC50 carries little information about drug efficacy, only about potency. We often prefer the AUC because it averages out the efficacy and potency (at the cost of being dependent on the chosen concentration range).

Best, Petr