Abstract
In cancer pharmacology, a drug candidate's therapeutic potential is typically expressed as its ability to suppress cell growth. Different methods in assessing the cell phenotype and calculating the drug effect have been established. However, inconsistencies in drug response outcomes have been reported, and it is still unclear whether and to what extent the choice of data post-processing methods is responsible for that. Studies that systematically examine these questions are rare. Here, we compare three established calculation methods on a collection of nine in vitro models of glioblastoma, exposed to a library of 231 clinical drugs. The therapeutic potential of the drugs is determined on the growth curves, using growth inhibition 50% (GI50) and point-of-departure (PoD) as the criteria. An effect is detected on 36% of the drugs when relying on GI50 and on 27% when using PoD. For the area under the curve (AUC), a threshold of 9.5 or 10 could be set to discriminate between the drugs with and without an effect. GI50, PoD, and AUC are highly correlated. The ranking of substances by different criteria varies somewhat, but the group of the top 20 substances according to one criterion typically includes 17-19 top candidates according to another. In addition to generating preclinical values with high clinical potential, we present off-target appreciation of top substance predictions by interrogating the drug response data of non-cancer cells in our calculation technology.