2021 年 20 巻 3 号 p. 106-111
We have developed a novel framework for inferring a chemical graph. Given an ANN that maps a feature vector x = f ( G ) of a chemical graph G to a predicted value η ( x ) of a chemical property to G, we formulate an integer program that simulates the functions f and η. Given a target value y*, we can infer a chemical graph G such that η ( f ( G ) ) = y * by solving the integer program.