Molecule superstructures for computer-aided molecular and process design

Abstract

Integrated molecular and process design optimizes process variables together with molecules as an additional degree of freedom. The integrated design needs to represent the molecule in a machine-readable way that can be operated on by an optimization algorithm. For this purpose, group-contribution methods have been established as property models in molecular design applications. The underlying molecular representation for a group-contribution method is the number of occurrences of different pre-defined groups within the molecule. However, this way of encoding a molecule omits information about the structure of the molecule and thus limits the molecular detail available during design. In this work, we present a graph-based molecular representation approach that encodes the full structure of the molecule during optimization. This approach unlocks additional higher-fidelity property prediction methods for integrated molecular and process design while still allowing the use of gradient-based optimization algorithms. The framework is applied in a case study that designs the working fluid for an organic Rankine cycle using the heterosegmented gc-PC-SAFT equation of state as property prediction model. The molecular superstructure representation is shown to enable the efficient integration of advanced property models into molecular design.