Published in: Journal, Article, Volume : 46, Issue : 3, Pages : 1394-1401
DOI : 10.1021/ci050459i
Author : Biswas, Dhrubajyoti; Roy, Sujata; Sen, Srikanta
Abstract : A knowledge-based simple score has been developed for indexing the oral druglikeness of compounds based on the concept that oral druglikeness should be independent of the drug targets and, thus, are closely related to the global absorption, distribution, metabolism, and excretion related properties. We have considered several simple mol. descriptors as the key determinants of druglikeness. The patterns of the distributions of these mol. descriptors for a set of drug mols. have been extracted using a nonlinear neural network method. We assumed direct correlations of these patterns to the expectation values that a given compound may behave like a drug. On the basis of this assumption, we have defined a simple druglike index or score (DLS) combining the contributions coming from the descriptors considered. This index scales the druglikeness of a compound in the range 0.0-1.0, 1.0 being the highest druglikeness. The index applied for a drug data set, a mixed data set, and three different bioactive databases produced expected features and indicated that even the marketed drugs have druglike scores varying over a considerable range. A total of 73.3% of the drugs considered showed DLS > 0.5, while it is only 44.7% for the HIC-Up compounds (unbiased ligand database). For the ChemBank, Asinex-Gold collection, and NCI databases 61.2%, 76.0%, and 79.1% of the compounds have DLS > 0.5.