Benchmarking the Reliability of QikProp. Correlation between Experimental and Predicted Values

Abstract The theoretical prediction power of the software package QikProp was tested. This was achieved by comparing experimentally known results to predicted values. First, simple molecular descriptors for physicochemical properties: octanol–water partition (log P ), water solubility (log S ), dipole moment, Ionisation Potential (IP) and Electron Affinity (EA) were compared to their experimentally determined counterparts. For all of the descriptors, except EA, a relatively good linear correlation was obtained. Experimentally derived EA values are relatively scarce and often quite inconsistent, which made it difficult to construct a reliable test collection. When compared to values calculated by the Density Functional Theory (DFT) method a reasonable correlation was observed but there is much room for improvement. A clear Gaussian distribution pattern was obtained when a collection of ∼470 marketed orally bioavailable drug compounds was used to generate the physicochemical properties investigated. The idea was explored whether the prediction power of ADME modules could be tested in a qualitative way, based on broad assumptions, using different classes of marketed drug compounds. It was found that this approach gives one a good idea about which modules deserve further attention for evaluation. In this way it is concluded that cell permeation and the blood–brain barrier modules merit more evaluation work whereas work on the HERG K + and CNS activity modules was discontinued.