D3SC: In Silico Design of Molecular Catalysts for C-H Functionalization via Machine Learning Algorithms

我们能否教会机器化学直觉，让机器帮助我们发现具有有用性质的新分子和材料？这样做能加速这些发现吗？这些都是田纳西大学的Vogiatzis博士正在解决的问题。为了实现这些目标，他正在应用能够学习新的复杂数学函数的统计模型。这些功能有能力通过寻找有助于重要和理想性质的模式，将有意义的化学信息与分子结构联系起来。这种化学驱动的机器学习（CDML）计算方法对于各种环境、生物和能源相关问题是一种很有前途的技术。为了证明CDML的强度和适用性，Vogiatzis博士和他的研究小组正在研究模拟含铁酶作用的铁物种的化学性质。Vogiatzis博士正在为学生提供跨学科的研究机会，包括那些来自弱势群体的学生。反过来，他们正在学习数据科学和机器学习方法，这是未来技术发展的重要工具。Vogiatizis博士的研究小组开发的计算工具将免费提供给其他对机器学习和化学感兴趣的研究人员。在化学催化项目和化学学部D3SC（数据驱动的化学发现科学）计划的资助下，田纳西大学的Vogiatzis博士正在开发计算工具，用于对大型分子复合物文库进行高效、高通量的计算筛选。长期目标是通过量子化学和机器学习设计下一代高效碳氢化合物功能化催化剂。他的研究小组目前正在研究一种综合计算方案，用于检查一类碳氢化合物活化的反应位点，但所提出的方法也可转移到其他化学过程中。目前研究的仿生催化位点是Fe(IV)-oxo中间体，它是血红素和非血红素酶的活性位点，之所以选择Fe(IV)-oxo中间体，是因为有大量的文献可以指导计算模型的发展。Vogiatzis博士正在吸引来自弱势群体的本科生和研究生参与他的研究。他还吸引来自田纳西州东部和中部的学生讲述他的科学兴趣。此外，开发免费的、开源的软件对于科学界和科学的进步是非常重要的。由化学催化项目资助开发的计算工具将免费提供给其他对机器学习和化学感兴趣的研究人员。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Can we teach a machine chemical intuition so that the machine can help us discover new molecules and materials with useful properties? Can doing so accelerate these discoveries? These are questions that Dr. Vogiatzis of the University of Tennessee is addressing. To achieve these goals, he is applying statistical models that can learn new complex mathematical functions. These functions have the power to connect meaningful chemical information with molecular structure by looking for patterns that contribute to important and desirable properties. This chemically-driven machine learning (CDML) computational approach is a promising technique for a large variety of environmental, biological, and energy-related problems. To demonstrate the strength and applicability of CDML, Dr. Vogiatzis and his research group are examining the chemical properties of iron species that mimic the action of iron-containing enzymes. Dr. Vogiatzis is providing interdisciplinary research opportunities to students, including those from underrepresented groups. In turn, they are learning about data science and machine learning methodologies, an important tool for the development of tomorrow's technologies. The computational tools developed in Dr. Vogiatizis' research group are being provided free to other researchers interested in machine learning and chemistry.With funding from the Chemical Catalysis Program and the D3SC (Data Driven Discovery Science in Chemistry) initiative of the Chemistry Division, Dr. Vogiatzis of the University of Tennessee is developing computational tools for efficient high-throughput computational screening of large libraries of molecular complexes. The long-term target is the design of the next generation of catalysts for efficient C-H functionalization via quantum chemistry and machine learning. His research group is currently working on an integrated computational protocol that examines one class of reactive sites for C-H activation, but the proposed methodology is transferable to other chemical procedures as well. The biomimetic catalytic site that is currently examined is the Fe(IV)-oxo intermediate, active site of heme and non-heme enzymes, and is chosen due to the vast literature that can guide the development of the computational model. Dr. Vogiatzis is engaging undergraduate and graduate students from underrepresented groups in his research. He is also engaging students from East and Central Tennessee about his scientific interests. In addition, the development of free, open-source software is of high importance for the scientific community and the advance of the science. The computational tools that are developed with funding from the Chemical Catalysis Program are being provided free-of-charge to other researchers interested in machine learning and chemistry.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.