A Scoring Scheme for Discriminating between Drugs and Nondrugs

Abstract

A scoring scheme for the rapid and automatic classification of molecules into drugs and nondrugs was developed. The method is a valuable new tool that can aid in the selection and prioritization of compounds from large compound collections for purchase or biological testing and that can replace a considerable amount of laborious manual work by a more unbiased approach. It is based on the extraction of knowledge from large databases of drugs and nondrugs. The method was set up by using atom type descriptors for encoding the molecular structures and by training a feedforward neural network for classifying the molecules. It was parametrized and validated by using large databases of drugs and nondrugs (169 331 molecules from the Available Chemicals Directory, ACD, and 38 416 molecules from the World Drug Index, WDI). The method revealed features in the molecular descriptors that either qualify or disqualify a molecule for being a drug and classified 83% of the ACD and 77% of the WDI adequately.

Classifying ‘Drug-likeness' with Kernel-Based Learning Methods

“…Byvatov et al considered the data set that has also been used in ref 14. However in the experiments by Byvatov et al, the WDI to ACD ratio has been skewed from the original 1:4.4 to almost 1:1.…”

“…We therefore were able to reduce the error rate for the task of classifying drugs (WDI) vs nondrugs (ACD) by more than 60% compared to Byvatov et al 12 and Sadowski and Kubinyi. 14 …”

Comparison of Ranking Methods for Virtual Screening in Lead-Discovery Programs

“…6 Finally, if the 3D structure of the biological target is known, then a docking study can be carried out to identify those database molecules that are complementary to the binding site. 7 This paper reports a comparison of methods that can be used for the third class of virtual screening methods, which includes such common approaches as substructural analysis, 8,9 genetic algorithms, 10 neural networks, 11,12 and decision trees. [13][14][15] Two main classes of approach are possible: ranking methods order a database in order of decreasing probability of activity and classification methods divide a database into those molecules that are predicted to be active and those that are predicted to be inactive.…”

Gradual in Silico Filtering for Druglike Substances

“…This procedure is in line with previous studies. 3,11,26 Classification of the compounds into drugs and nondrugs was performed under consideration of the following criteria:…”

“…Further computed descriptors that were also available to the classification algorithms, but not used for clustering, comprised the number of hydrogen-bond donors and hydrogenbond acceptors, the total number of halogen atoms, the number of carboxylic acids as well as other acids, the respective count of hydroxy, amino, NO 2 , sulfoxy, sulfonyl, SO 3 , nitrile, CF 3 , CCl 3 , and ester groups as well as the total sum of these functional groups, the number of unsuitable groups according to Flower 32 also including occurrences of matching substructures according to the SMARTS strings compiled from Hann et al, 31 Oprea, 29 Rishton, 43 and Anzali et al 13 (see Table S2 in the Supporting Information), the respective number of 3-, 4-, 5-, and 6-membered rings as well as aromatic 5-and 6-membered rings, the total number of rings, the number of violations of Lipinski's rule, 28 the 50% and 80% criteria of drugs by Ghose, Viswanadhan, and Wendoloski, 30 Oprea's druglike criteria at 70%, 29 and Hutter's druglikeliness index. 6 Quantum chemical descriptors were obtained from semiempirical AM1 calculations using a modified version of the program package VAMP.…”