Multiple objective resource allocation in product and process development

Abstract

A comprehensive hierarchical methodology has been developed to assist decision-makers allocate resources for experimentation during the initial-tages of pharmaceutical and chemical process development. The goal is to identify the most useful information that can be obtained for the least amount of effort and time. The allocation of resources for information gathering is based on Bayesian experimental design. Specifically, experimental designs for parameter estimation, model discrimination, and decisionmaking have been examined. Solving some of these design problems rigorously has not previously been attempted due to the mathematical complexity involved and sheer computational intensity of classical methods. The enabling technology is the use of polynomial chaos expansions to represent process and decisions models. A compact representation of uncertainty permits a rapid evaluation of expected values and variances in the decision models. In typical applications the computational burden was reduced by more than four orders of magnitude. The technique allows processes with industrial levels of complexity to be analyzed. The methodology takes a hierarchical approach. Initially the process subsystem that most adversely affects the objectives is identified. In this way resources are only allocated to studying the most important components. Metrics for measuring financial, environmental, and safety, objectives at different stages of the development process are suggested. The performance measures are unique to pharmaceutical and chemical manufacturing; however, the mathematical techniques developed are universally relevant. Examples showcase the experimental design approaches, the performance metrics, and the hierarchical modeling. A comprehensive case study, production of recombinant heparinase, highlights the most important aspects for an industrially relevant process.