Machine learning: NMR Shift and J-Coupling Prediction

机器学习：NMR 位移和 J 耦合预测

• 批准号: 2444168
• 金额: --
• 依托单位: University of Bristol
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2020
• 资助国家: 英国
• 起止时间: 2020 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2444168
• 关键词: Machine; learning; NMR; Shift; Coupling

Though there has been an increase in interest in the application of machine learning models into assisting drug development, little focus has been placed into applying these machine learning techniques to improve NMR property prediction. By developing a model that is capable of predicting these parameters, the process of elucidating molecules can be rapidly accelerated. IMPRESSION (Intelligent Machine PREdiction of Shift and Scalar information of Nuclei) was a successful attempt at predicting NMR parameters from a 3D molecular structure comparable to the accuracy of computationally intensive quantum mechanical calculations.By building upon the foundational IMPRESSION model, the capabilities of it could be further expanded by: predicting a wider range of molecules through a more diverse training dataset, increasing the accuracy of the underlying DFT method used to generate the training data, expanding the parameter prediction into physical properties, and including multi-conformer analysis. These developments could result in a highly robust chemical toolkit for structure elucidation.

尽管人们对应用机器学习模型辅助药物开发的兴趣日益浓厚，但很少有人关注应用这些机器学习技术来改进 NMR 特性预测。通过开发能够预测这些参数的模型，可以迅速加速阐明分子的过程。 IMPRESSION（原子核位移和标量信息的智能机器预测）是从 3D 分子结构预测 NMR 参数的成功尝试，其精度可与计算密集型量子力学计算的精度相媲美。通过构建基础 IMPRESSION 模型，它的功能可以进一步扩展：通过更多样化的训练数据集预测更广泛的分子，提高用于预测的基础 DFT 方法的准确性。 生成训练数据，将参数预测扩展到物理属性，并包括多构象分析。这些进展可能会产生用于结构阐明的高度稳健的化学工具包。