Exploring Molecular Context: Regioselectivity Prediction

第一年：一般技能培训和实验室轮换2-4年：该项目旨在开发和实施一个通用的工作流程，基于分子背景的概念，用于反应预测。作为一个初步的案例研究，我们将选择芳族氯化反应，其中的起始材料的结构起着决定区域选择性的主要作用。目前的芳烃取代模型是从溴化反应中发展起来的，尽管它们的适用范围非常重要，但尚未得到广泛的研究。本案例研究将从数据收集和数据清理开始;因此，我们将选择一个很好理解的反应，并提供各种实验结果。下一步将是分析当前模型和不同预测模型无法给出准确结果的群体情况。我们希望分析模型的局限性，以指导隐藏的缺失功能的分子背景下的研究。第二个案例研究将评价吲唑的反应性，其中反应性和反应条件起重要作用。这项额外的研究将补充反应预测的概括，因为它必须理解反应物结构以外的信息。最后，将使用计算化学和统计学对分子参数化进行深入研究，以获得反应预测的一般方法。因此，我们计划为反应预测反应开发一个通用工作流程，可以突出最重要的分子和实验特征，并使用系统的分子背景概念改进化学中机器学习模型的开发。

Year 1: General skills training and lab rotationsYears 2-4: The project aims to develop and implement a general workflow, based on the concept of a molecular context, for reaction prediction. As an initial case study, we will select aromatic chlorination reactions, where the structure of the starting material plays the major role on defining regioselectivity. Current models for aromatic substitution are developed from bromination and their range of applicability is not widely explored, although it is of paramount importance. This case study will start with data collection and data cleaning; thus, we will choose a reaction well understood and with various experimental results available. The next step will be the analysis of current models and group situations where different prediction models fail to give an accurate result. We expect the analysis of the limitations of models to guide the study of hidden missing features for the molecular context. A second case study will be evaluated on the reactivity of indazoles, where reactivity and reaction conditions play important roles. This additional study will complement a generalisation of reaction prediction because it must comprehend information beyond reactant structure. Finally, an in-depth study of molecular parametrisation - using computational chemistry and statistics - will be conducted to have a general approach towards reaction prediction. Therefore, we plan to develop a general workflow for reaction prediction reactions that can highlight the most important molecular and experimental features for the transformation and improve the development of machine learning models in chemistry using the concept of a molecular context of the system.