Multiobjective Pareto optimisation of pharmaceutical product formulation using radial basis function network and non-dominated sorting differential evolution

Abstract

Pharmaceutical product formulation often involves several composition factors and response characteristics and its optimal formulation requires the best combination of formulation variables to satisfy the multiple and conflicting characteristics of the product. In this work, a novel multiobjective Pareto optimisation strategy is developed by combining a radial basis function network (RBFN) with a non-dominated sorting differential evolution (NSDE) and applied for optimal formulation of a trapidil product involving conflicting response characteristics. The RBFN models of this strategy are developed by using spherical central composite design data of trapidil formulation variables representing the amounts of microcrystalline cellulose, hydroxypropyl methylcellulose and compression pressure and the corresponding response data of release order and rate constant. These models in combination with NSDE are augmented with naive and slow and constraint techniques to generate Pareto optimal solutions for product formulation. The optimal formulation results of RBFN-NSDE are compared with those of multiple regression model-based evolutionary Pareto optimisation strategy. The RBFN-NSDE is found to exhibit better Pareto optimal performance for pharmaceutical product formulation.