(173g) Multi Objective Molecular Design of Reactants and Products

Authors: 
Dev, V., Auburn University
Chemmangattuvalappil, N. G., University of Nottingham - Malaysia
Eden, M. R., Auburn University

In reactive systems, computer-aided molecular design (CAMD) is performed to design solvents that enhance reaction kinetics, improve catalyst activity/lifetime as well as reactants and products. CAMD of reactants and products helps in their selection such that target properties of interest are optimized with satisfaction of property constraints. Owing to the scarcity of CAMD methodologies that design reactants and products, there has been an increased interest to address this issue. The few recently available methodologies to design reactants and products however have their shortcomings. The earliest available methodologies are restricted to cases where only structures of a single reactant and product are variable. Also, employment of linear property models and lack of a methodology to treat property constraints have been some of the other restrictions.  Dev et al. [1, 2] recently developed two algorithms to design reactants and products which are unrestricted by the number of reactants and products whose structures are uknown. They also made provisions for non-linear property models and property constraints. In their first algorithm [1], only the dominant properties of the products are optimized. Also, only the products are subjected to their individual set of property constraints. In their second algorithm [2], properties that are dependent on the structures of both reactants and products are optimized. Also, all reactants and products are subjected to their respective set of property constraints. In this work, we have developed an algorithm that generates structures of reactants and products irrespective of their numbers such that each reactant and product’s target property(s) is to be optimized. Also, each reactant and product is subjected to a set of property constraints. All properties have been expressed in terms of molecular structure using different types of property models. Since there exists a need to treat different types of property models on a single platform, signature descriptors have been utilized. A variety of property models can be expressed in terms of signature descriptors which are molecular building blocks. They enable the simultaneous utilization of Quantitative Structure Activity/Property Relationships (QSARs/QSPRs) and Group Contribution Models (GCMs) [3]. In the formulated problem, all property models are expressed in terms of the occurrence number of signature descriptors. Relationships have been developed that relate the occurrence number of signatures of reactants and products. This is achieved by tracking the atoms in the reactants to the products by taking into consideration the reaction mechanism. Thus the reactants and products have been structurally related. Due to this structural relation, property constraints acting on reactants and products influence the selection of both optimal reactants and products. The objective functions of each of the reactants and products display a similar influence. Since the selection of each reactant and product is dependent on the property constraints and property objective functions of the other reactants and products in addition to its own, the problem is formulated as a multi-objective optimization problem when objectives are conflicting. This problem is solved using the Augmented Epsilon-Constraint Method (AUGMECON) developed by Mavrotas [4]. AUGMECON generates pareto optimal solutions while completely avoiding weakly pareto optimal solutions. The generated solutions also include unsupported pareto optimal solutions. Specifically for multi-objective integer programming (MOIP) problems, an improved version of AUGMECON, AUGMECON2 [5] can also be utilized. AUGMECON2 is very efficient in generating the entire pareto optimal set for MOIP problems compared to AUGMECON and some earlier available methods. In this contribution, the utility and novelty of the developed methodology has been highlighted through a case study.

[1] V.A. Dev, N.G. Chemmangattuvalappil, M.R. Eden, 2014, Computer Aided Chemical Engineering, 34, 291-296

[2] V.A. Dev, N.G. Chemmangattuvalappil, M.R. Eden, 2015, Computer Aided Chemical Engineering, 37, 1445-1450

[3] N.G. Chemmangattuvalappil, C.C. Solvason, S. Bommareddy, M.R. Eden, 2010, Computers & Chemical Engineering, 34, 12, 2062-2071

[4] G. Mavrotas, 2009, Applied Mathematics and Computation, 213, 455-465

[5] G. Mavrotas, K. Florios, 2013, Applied Mathematics and Computation, 219, 9652-9669