Development of Computer-Aided Molecular Design Methods for Bioengineering Applications.

摘要

Computer-aided molecular design (CAMD) offers a methodology for rational product design. The CAMD procedure consists of pre-design, design and post-design phases. CAMD was used to address two bioengineering problems: design of excipients for lyophilized protein formulations and design of ionic liquids for use in bioseparations. Protein stability remains a major concern during protein drug development. Lyophilization, or freeze-drying, is often sought to improve chemical stability. However, lyophilization can result in protein aggregation. Excipients, or additives, are included to stabilize proteins in lyophilized formulations. CAMD was used to rationally select or design excipients for lyophilized protein formulations. The use of solvents to aid separation is common in chemical processes. Ionic liquids offer a class of molecules with tunable properties that can be altered to find optimal solvents for a given application. CAMD was used to design ionic liquids for extractive distillation and in situ extractive fermentation processes.;The pre-design phase involves experimental data gathering and problem formulation. When available, data was obtained from literature sources. For excipient design, data of percent protein monomer remaining post-lyophilization was measured for a variety of protein-excipient combinations. In problem formulation, the objective was to minimize the difference between the properties of the designed molecule and the target property values. Problem formulations resulted in either mixed-integer linear programs (MILPs) or mixed-integer non-linear programs (MINLPs).;The design phase consists of the forward problem and the reverse problem. In the forward problem, linear quantitative structure-property relationships (QSPRs) were developed using connectivity indices. Chiral connectivity indices were used for excipient property models to improve fit and incorporate three-dimensional structural information. Descriptor selection methods were employed to find models that minimized Mallow's Cp statistic, obtaining models with good fit while avoiding overfitting. Cross-validation was performed to access predictive capabilities. Model development was also performed to develop group contribution models and non-linear QSPRs. A UNIFAC model was developed to predict the thermodynamic properties of ionic liquids.;In the reverse problem of the design phase, molecules were proposed with optimal property values. Deterministic methods were used to design ionic liquids entrainers for azeotropic distillation. Tabu search, a stochastic optimization method, was applied to both ionic liquid and excipient design to provide novel molecular candidates. Tabu search was also compared to a genetic algorithm for CAMD applications. Tuning was performed using a test case to determine parameter values for both methods. After tuning, both stochastic methods were used with design cases to provide optimal excipient stabilizers for lyophilized protein formulations. Results suggested that the genetic algorithm provided a faster time to solution while the tabu search provides quality solutions more consistently.;The post-design phase provides solution analysis and verification. Process simulation was used to evaluate the energy requirements of azeotropic separations using designed ionic liquids. Results demonstrated that less energy was required than processes using conventional entrainers or ionic liquids that were not optimally designed. Molecular simulation was used to guide protein formulation design and may prove to be a useful tool in post-design verification. Finally, prediction intervals were used for properties predicted from linear QSPRs to quantify the prediction error in the CAMD solutions. Overlapping prediction intervals indicate solutions with statistically similar property values. Prediction interval analysis showed that tabu search returns many results with statistically similar property values in the design of carbohydrate glass formers for lyophilized protein formulations. The best solutions from tabu search and the genetic algorithm were shown to be statistically similar for all design cases considered. Overall the CAMD method developed here provides a comprehensive framework for the design of novel molecules for bioengineering approaches.