Electrotopological State Indices to Assess Molecular and Absorption, Distribution, Metabolism, Excretion, and Toxicity Properties

Abstract

This article examines the role of topological structure descriptors as in silico tools for modeling ADMET properties. Topological structure descriptors represent significant features of molecular structure by encoding in numerical form basic attributes of atoms in molecules. This approach is a broadly based view of molecular structure as it relates to intermolecular interactions, and is known under the name structure information representation. The basis for the structure information representation is the view of a molecule as an information network. The molecule is much more than a collection of geometric points with assigned property values. Two structure information representation methods are described, the electrotopological state (E-state) and molecular connectivity. Both methods are based on basic properties of each atom in the molecule, counts of electrons in the valence state and partitioned among π orbitals and lone pairs. These properties are used to create a measure of the electron accessibility at each atom in a molecule. This concept of electron accessibility has been extended to the creation of E-state values for all atoms of the same type in a molecule and also for bond types. The other structure representation method is a representation of the whole molecule, known as molecular connectivity, in which whole-molecule structure attributes are encoded, such as degree of skeletal branching, degree of branching adjacency, ring structure, and molecular size. The combination of E-state descriptors that focus on individual atoms together with molecular connectivity indices that encode whole-molecule features has proven powerful in ADMET modeling. When the whole range of topological descriptors is considered, this approach is known as structure information representation. One major feature of this approach is that the model yields information directly related to the structure attributes that are significant for the property under investigation, leading to information that is useful in the structure modification process in drug design. The structure information representation approach together with nonlinear modeling techniques such as artificial neural networks are described for several ADMET properties. In each case, some of the significant aspects of the study are presented. The wide range of properties and the results indicate the power and usefulness of the in silico tools described in this article.