Molecular Descriptors for Drugs: A Discriminant Analysis

Abstract

Background: The biological activity of a compound is assumed to be encoded in its chemical composition and geometric structure, from which physico-chemical, electrotopological, and graph theory-derived properties can be determined.

Objective: This study aimed to identify molecular descriptors derived from Dragon® 6 software that can
discriminate compounds as drug or nondrug.

Methodology: In this study, over 4000 molecular properties were obtained for approximately 2000 known
drugs and 2000 nondrugs on which Linear Discriminant Analysis was performed.

Results: Compounds can be discriminated between drug and nondrug with 81% accuracy using only two
molecular descriptors, the information index HVcpx and the topological index MDDD.

Conclusion: A “Rule of Three” (HVcpx ≤ 3 and MDDD ≥ 30) seems to confer druglikeness in compounds. This rule can be used as additional filter in high throughput screening of compounds in any drug discovery research.