A Novel Method for the Inverse QSAR/QSPR based on Artificial Neural Networks and Mixed Integer Linear Programming with Guaranteed Admissibility

Inverse QSAR/QSPR is a well-known approach for computer-aided drug design. In this study, we propose a novel method for inverse QSAR/QSPR using artificial neural network (ANNs) and mixed integer linear programming In this method, we introduce a feature function f that converts each chemical compound G into a vector f (G) of several descriptors of G. Next, given a set of chemical compounds along with their chemical properties, we construct a prediction function iv with an ANN so that psi( f (G)) takes a value nearly equal to a given chemical property for many chemical compounds G in the set. Then, given a target value y* of the chemical property, we conversely infer a chemical structure G K having the desired property y* in the following way. We formulate the problem of finding a vector x* such that (i) psi(x*) = y and (ii) there exists a chemical compound G* such that f(G*) = x* (if one exists over all vectors x* in (i)) as a mixed integer linear programming problem (MILP). In an existing method for the inverse QSAR/QSPR, the second condition (ii) was not guaranteed. For acyclic chemical compounds and some chemical properties such as heat of formation, boiling point, and retention time, we conducted computational experiments.