Screening molecules for their drug-like index


Anwar Rayan1, Andrea Scaiewicz2, Inbal Geva-Dotan2, Dinorah Barasch1, and Amiram Goldblum1. (1) Medicinal Chemistry and Natural Products, Hebrew University of Jerusalem, School of Pharmacy, Jerusalem, 91120, Israel, (2) Department of Medicinal Chemistry and the David R. Bloom Center for Pharmacy, Hebrew University of Jerusalem, Jerusalem Israel, 91120, Israel
A new drug like index (DLI) is presented. It is formed by applying the Iterative Stochastic Elimination (ISE) algorithm (1-4) for constructing a set of options to differentiate between drugs and non-drugs (CMC/ACD) with appropriate training and test sets. The set of best solutions forms the basis for constructing DLI, as a sum over the relative contributions of true and false negatives and positives to each of the solutions. The best k-descriptor combinations out of some 150 descriptors have been picked by ISE, as well as their optimal limits for differentiating between drugs and non-drugs. DLI has been applied to several groups of the MDDR database, resulting in several implications for DLI values of different phases in clinical trials. The use of DLI for constructing combinatorial libraries will be demonstrated.


(1) Glick et al., PNAS 2002, 99, 703-708. (2) Glick et al., Proteins 2000, 38, 273-287. (3) Rayan et al., Curr Med Chem 2004, 11, 675-692. (4) Rayan et al., J Mol Graph Model 2004, 22, 319-333.