Evolutionary algorithms for automated drug design towards target molecule properties

This paper presents an evolutionary algorithm for the automated design of molecules that could be used as drugs. It is designed to provide the medicinal chemist with a number of candidate molecules that comply to pre-defined properties. These candidate molecules can be promising for further evaluation. The proposed algorithm is implemented as an extension to the so-called Molecule Evoluator [3] which implements an interactive evolutionary algorithm. The Molecule Evoluator is extended with an automated evolutionary algorithm that implements a variable sized population and bases its search on target-bounds that are set for a number of molecule properties. Moreover, the algorithm uses a selection procedure based on the notion of Pareto domination. The results show that it is indeed possible to apply the concept of evolutionary computation on automated molecule design using target-bounds for molecule properties as optimization goals. For practical usage, the presented algorithm could serve as a starting point, but should be further improved with respect to diversity within the generated set of molecules.