CDS&E: Collaborative Research: Towards computational discovery of synthetically feasible porous organic frameworks

CDS

• 批准号: 1953245
• 负责人: Srinivas Rangarajan
• 金额: $ 42万
• 依托单位: Lehigh University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-09-01 至 2024-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1953245&HistoricalAwards=false
• 关键词: CDS& Collaborative; Research; Towards; computational

Covalent organic frameworks (COFs) belong to an emerging class of organic materials with low density, high thermal stability, and stable porosity, with applications including molecular separations, catalysis, energy storage, semiconductors, drug delivery, and single-molecule sensors. They are constructed via dynamic covalent assembly of organic building-block (BB) molecules, a process wherein the BBs essentially polymerize to form an extended crystalline material. The COF “universe,” i.e. the set of all possible COF structures, is vast and potentially comprises billions of potential candidates with varying pore size and geometry, chemical composition, and functionality. Only a small fraction of this “universe” has been synthesized so far, although new COFs are continuously being reported. Nevertheless, there are still very few reported COFs with more than two types of pores, a critical shortcoming because COFs with combinations of pore sizes hold promise for enhancing catalysis under confinement, molecular separation processes, and gas storage applications. A prominent idea for designed synthesis of COFs, known as reticular chemistry, is to identify rigid BB molecules that can be uniquely assembled into the desired COF pattern via covalent bonds. While extremely useful, the related chemical design is often implemented manually, and is therefore not scalable to more complex topologies nor is it able to enumerate the vast space of COFs. In this context, the goal of this research is to develop a computational method to automatically generate synthetically feasible 2D covalent organic frameworks with multiple pore sizes and thereby guide experimental efforts to synthesize complex structures. Synthetic feasibility considers whether the building block(s) can be easily created using known chemistries and starting materials and easily assembled into the requisite crystalline COF structure.This research brings together tools from cheminformatics, reaction network generation, advanced molecular simulations, and artificial intelligence to create an automated method to identify COFs that can be synthesized using easily available starting materials and proven organic chemistries. This method will be used to create a database of COFs with complex (in particular heteroporous) topology. Given a target structure, the method will identify the necessary building block structure and its chemical functionality via coarse-grained molecule-like patchy particle simulations. The resulting information will be used to generate potential molecular building blocks using a reinforcement learning-based biased automated network generation process. The synthetic complexity of these molecular building blocks will then be evaluated using cheminformatics tools and algorithms. For the most synthetically feasible molecules, their synthesis routes will be generated using the concept of retrosynthesis via automated network generation. The end result of this process will be a list of theoretically determined synthetically feasible COFs, their building blocks, and their synthesis routes. Such lists will be compiled for each tiling and ranked based on the synthesis scores. A few of the most promising COFs identified through this strategy will be verified experimentally in a bottom-up assembly involving synthesis of the building blocks and their assembly using orthogonal reaction chemistries. To integrate research and education, the PIs will employ the concept of student-led creation of original scientific research content as part of curricular training, or “class sourcing,” by designing course projects wherein the cumulative expertise of the entire cohort of students is leveraged to identify strategies to synthesize new building blocks and thereby improve rules for network generation.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.

共价有机骨架(COF)是一类密度低、热稳定性高、孔隙率稳定的有机材料，在分子分离、催化、储能、半导体、药物输送和单分子传感器等领域具有广泛的应用前景。它们是通过有机积木(BB)分子的动态共价组装构建的，在这个过程中，BBS基本上聚合形成延伸的晶体材料。COF“宇宙”，即所有可能的COF结构的集合，是巨大的，潜在地包括数十亿个潜在的候选者，具有不同的孔大小和几何形状、化学成分和功能。到目前为止，虽然不断有新的COF被报道，但到目前为止，只合成了这个“宇宙”的一小部分。然而，仍然很少有报道的COF具有两种以上的孔类型，这是一个关键的缺陷，因为具有不同孔径组合的COF有望在受限条件下增强催化、分子分离过程和气体存储应用。设计合成COF的一个突出的想法是识别刚性BB分子，这些分子可以通过共价键唯一地组装成所需的COF模式。虽然非常有用，但相关的化学设计往往是手动实现的，因此不能扩展到更复杂的拓扑，也无法列举COF的巨大空间。在这种背景下，本研究的目标是开发一种计算方法来自动生成具有多种孔径的可合成的2D共价有机骨架，从而指导合成复杂结构的实验努力。合成可行性考虑的是构筑块(S)是否可以使用已知的化学成分和起始材料轻松地创建并组装成必要的结晶COF结构。这项研究结合了化学信息学、反应网络生成、高级分子模拟和人工智能的工具，创建了一种自动鉴定COF的方法，可以使用容易获得的原料和成熟的有机化学来合成COF。该方法将用于创建具有复杂(特别是杂多的)拓扑的COF的数据库。在给定目标结构的情况下，该方法将通过粗粒度的分子状片状粒子模拟来确定必要的构建块结构及其化学功能。所得到的信息将被用于使用基于强化学习的有偏见的自动网络生成过程来生成潜在的分子构建块。然后将使用化学信息学工具和算法来评估这些分子构建块的合成复杂性。对于最具合成可行性的分子，其合成路线将通过自动网络生成的逆合成概念生成。这一过程的最终结果将是一份理论上确定的可综合可行的COF的清单、它们的构件和它们的合成路线。这样的名单将针对每一块瓷砖进行汇编，并根据综合分数进行排名。通过这一策略确定的一些最有希望的COF将在自下而上的组装中得到实验验证，该组装涉及构建块的合成及其使用正交反应化学反应的组装。为了整合研究和教育，PIS将采用学生主导创造原创科学研究内容的概念作为课程培训的一部分，或通过设计课程项目，利用整个学生群体的累积专业知识来确定策略，以合成新的构建块，从而改进网络生成规则。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

A conceptual study of transfer learning with linear models for data-driven property prediction

• DOI: 10.1016/j.compchemeng.2021.107599
• 发表时间: 2021-11
• 期刊: Comput. Chem. Eng.
• 作者: Bowen Li;S. Rangarajan

Towards a chemistry-informed paradigm for designing molecules

• DOI: 10.1016/j.coche.2021.100717
• 发表时间: 2022
• 期刊: Current opinion in chemical engineering
• 影响因子: 6.6
• 作者: S. Rangarajan