I have a problem with the concept of false discovery rate. Assuming I have a panel of 50 anticancer drugs for which their IC50 values are reported for 30 cell lines. On the other hand, I have a modeling framework which allows me to predict IC50 for those drugs. I use Spearman correlations here (so basically, one p-value for each drug); the question is: should I use FDR to correct p-values? What does it mean when I use FDR here or not?
The experimental data is something like this:
Drug name | IC50 value for cell line #1 | cell line #2 | ..... |cell line #30
M1 | 0.7|0.53|...|0.65
M2 | 0.23|0.3|...|0.37
...
M50 |1.54|1.63|...|1.35
My modeling (mathematical) approach resulted in a set of IC50 (similar to the above results, with the difference that these results are simulated/predicted and not experimental). To compare the simulation results with experimental ones, I use Spearman correlations:
Drug name | Spearman p-value
M1 | p1
M2 | p2
...
M50 | p50
So, should I correct for multiple hypothesis?
Thanks
Dear Stephen,
Thanks for your response. The problem is not finding correlation across cell lines (here 30 cell lines). The problem is: I have a modeling framework which predicts the IC50 for all 50 drugs in those 30 cell lines. Therefore, there are two panels of 50(drug)X30(cell line), first panel is taken from experimental results and second is my predictions. So, to check the predictive power of my method, I calculate Spearman correlation between the prediction results and experimental results (each p-value for each drug). Finally, a set of 50 p-values (corresponding to drugs) will be generated. So, my question is, should I correct for multiple hypothesis testing or not? If yes, why?
Thanks
Thanks, yest it seems I need to use FDR. I came across this question because I see many papers with such type of comparisons/validations; however, they do not correct for multiple hypothesis. So, I was in doubt about to use or not to use!
Thanks again for your help.
- Badges
- Science topic