In V18 onwards (May 2016), the IC50 may have duplicated columns for a given drug. So some drugs are clustered together. The algorithm was implemented in GDSCTools in IC50Cluster class. It should be used as follow.
IC50 must be read with the IC50Cluster class as follows:
from gdsctools import *
ic50 = IC50Cluster("v18_data")
This will ensure also that the drug identifiers are unique. Indeed, in v18 data sets, columns for a given DRUG ID may be duplicated (for different drug concentration).
Then, as usual:
an = ANOVA(ic50, "GF.csv", "DRUG_DECODE.csv")
All default parameters were used except for the FDR threshold:
report.settings.FDR_threshold = 35