CAREER: Optimization Under Property Prediction Uncertainity in Molecular Design

这个项目的目的是设计分子产品，最优地满足一组设计目标，同时考虑到所采用的结构-性质关系的不准确性。现有的计算机辅助分子设计方法存在两个局限性：(1)取决于初始化和采用的搜索策略，它们可能收敛或不收敛到最佳分子设计；(2)即使偶然发现了数学上最好的分子设计，由于结构-性质表达式的不确定性，仍然不清楚是否真的存在最好的分子。本研究旨在克服这些缺点。为此，将利用机会约束规划、统计分析、混合整数线性和非线性规划算法以及确定性全局优化的混合方法。通过消除无意识地收敛于次优分子设计的陷阱，并量化性质预测不确定性对获得的分子设计的影响，PI希望提高分子产物识别的机会并加快识别速度。这将使实验工作只集中在最有希望的分子候选产物上。这项职业补助金计划的教育活动有两个方面：(a)课程开发，以及(b)学生指导、建议和研究介绍。利用计划的分子设计优化研究和先前在分子结构鉴定方面的工作，计划在两年内开设一门分子设计课程（宾夕法尼亚州立大学目前没有这门课程）。一门关于先进工艺合成和优化的课程将很快提交给教务委员会批准，以填补宾夕法尼亚州立大学化学工程课程中先进工艺合成课程的空缺。该课程将遵循夏季学期的初步版本，并将大量借鉴PI在优化方面的研究工作。

Abstract - Maranas - 9701771 The purpose of this project is to design molecular products which optimally meet a set of design objectives while accounting for inaccuracies in the employed structure-property relations. Existing approaches for computer-aided molecular design suffer from two limitations: (1) they may or may not converge to the best molecular design depending on their initialization and the adopted search strategy; and (2) even if by chance the mathematically best molecular design is found, it is still unclear if the best molecular is truly at hand due to the uncertainty in the structure-property expressions. This research aims at overcoming these shortcoming. To this end, a blend of chance-constrained programming, statistical analysis, mixed-integer linear and nonlinear programming algorithms, and deterministic global optimization will be utilized. By eliminating the pitfall of unknowingly converging to suboptimal molecular designs and quantifying the effect of property prediction uncertainty on obtained molecular designs, the PI expects to improve the chances and expedite the identification of molecular products. This will permit experimental efforts to be concentrated on only the most promising molecular candidate products. The educational initiatives planned in this CAREER grant are two-fold: (a) course development, and (b) student mentoring, advising and introduction to research. By taking advantage of the planned research on optimization in molecular design and prior work on molecular structure identification, a course on molecular design (which currently does not exist at Penn State) is planned to be offered in two years. A course on advanced process synthesis and optimization will shortly be submitted for approval by the Faculty Senate to fill a vacancy in advanced process synthesis courses in the Chemical Engineering curriculum at Penn State. The course will follow a preliminary version taught over the summer semester and will borrow heavily from the PI's resea rch work in optimization.