Evolutionary Algorithm for De Novo Molecular Design Considering Multi-Dimensional Constraints

The area of computer-aided molecular design has benefited from significant advances in technology and methodologies through consideration of problems with increasing complexity and size. The development of increasingly accurate and predictive property models, through application of powerful variable selection and improved mapping techniques, requires new approaches for designing structures meeting property requirements as informed by these models. Additionally, the long-standing desire to model and search an increasingly larger region of chemical space maintains. As such, the necessity of developing algorithms for solving these complex property models, most often containing molecular descriptors of varying type (e.g. information theoretic, charge based, constitutional) and dimensionality (e.g. 1D, 2D, 3D), within a large search space is realized. One such approach involves the implementation of guided stochastic algorithms, such as an evolutionary algorithm. This contribution outlines an evolutionary algorithm for solving computer-aided molecular design problems, with multi-dimensional criteria, in terms of spatial signature descriptors. The effect that various user defined variables have on this algorithm, and the resultant solutions, will be considered in further detail.