"G-SELC: A New Global Optimization Technique Using Genetic Algorithms, Tabu Search and Gaussian Processes"

该奖项是根据2009年美国复苏和再投资法案（公法111-5）资助的。在这项工作中，研究者开发了一种新的全局优化技术，这主要是由药物发现的应用驱动的。虽然鉴定有用的化合物对提高药物发现效率至关重要，但制药工业通常采用特殊方法来鉴定有前途的化合物。这项提议的研究旨在开发一种名为G-SELC的高效技术，以加速这一过程。为此，研究者开发了一个全局优化程序，使用局部搜索技术，如遗传算法和禁忌搜索，结合涉及高斯过程的统计建模。本文的研究也扩展到高斯过程中的分类变量。此外，研究者开发了有效的数值技术，以减少这批顺序优化问题的计算负担。从大量可行化合物中识别有前景的化合物是制药行业的一个重要而又具有挑战性的问题。拟议的研究有助于减少药物发现早期阶段的支出，从而创造显着的经济和社会效益。这种优化技术也被用于识别许多其他科学研究问题的最优解，如计算机实验、功能磁共振成像和纳米技术。研究者表明，在某些应用中，分类变量可以被视为连续的。这大大简化了高斯过程建模的计算。它不仅在药物发现方面具有深远的影响，而且在复杂的计算机建模中也具有深远的影响，其中广泛使用高斯过程建模，包括空气质量建模，脑血流计算模型的校准，预测气候和天气，颗粒流的统计力学，陆地模型，传染病动力学等。

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).In this work, the investigator develops a new global optimization technique, which is primarily motivated by applications in drug discovery. Although identification of useful compounds is critical to improving efficiency in drug discovery, pharmaceutical industries generally adopt ad hoc approaches to identify promising compounds. The proposed research aims to develop an efficient technique named G-SELC, which expedites this process. To this end, the investigator develops a global optimization procedure using local search techniques such as Genetic Algorithms and Tabu Search combined with statistical modeling involving Gaussian processes. This research is also extended to categorical variables in Gaussian process. In addition, the investigator develops efficient numerical techniques to reduce computational burden for this batch sequential optimization problem.Identifying promising compounds from a vast collection of feasible compounds is an important and yet challenging problem in pharmaceutical industry. The proposed research helps reduce the expenditure at the early stages of drug discovery, thus creating significant economic and social benefits. This optimization technique is also used to identify optimal solutions in many other scientific research problems such as computer experiments, functional magnetic resonance imaging and nanotechnology. The investigator shows that in some applications, categorical variables can be treated as continuous. This simplifies the computation in Gaussian process modeling significantly. It has far-reaching consequences not only in drug discovery, but also in complex computer modeling where Gaussian process modeling is used extensively which includes modeling air quality, calibration of computational models of cerebral blood flow, predicting climate and weather, statistical mechanics of granular flow, terrestrial models, dynamics of infectious diseases and so on.