Quantum Chemical Molecular Representations for Machine Learning

该项目旨在基于高效紧密结合（TB）量子化学（“量子特征”）为机器学习（ML）开发新的分子表征，并将这些表征连接到各种新的网络架构。这些模型将被应用于预测药物类分子的化学相关性质，如构象和变异构能、pKa值、溶解度或分配系数。这是一个世界领先的理论化学集团的一个项目，旨在开发和应用简化量子化学（QC）方法，并得到了科技公司默克的大力支持，默克在利用广泛的化学数据方面具有既定的能力。为了计算量子特征，将开发一个扩展AO基集（vDZP）中的新模型哈密顿量（ShellQ），该模型能够从参考DFT计算中精确地重现各种性质（原子电荷，壳族，键序，偶极矩，极化率），并且仍然普遍适用于包括有机金属体系在内的整个元素周期表。它首次在半经验的背景下解释了轨道收缩和电子极化等基本物理效应。它结合已建立的连续溶剂化理论来模拟溶剂化分子。该建议的进一步主要方面是基于ShellQ特性的神经网络架构的优化，特征表示的开发，分子训练数据集的自动化生成，以及受图像识别算法启发的最先进的多任务学习。一般来说，我们遵循Delta-ML策略，其中基于可用特征的快速QC计算（通常是已建立的GFN-xTB或GFN-FF方法）的修正项由网络计算。这整个方法应该为潜在的广泛的化学性质提供效率和准确性。

This project aims to develop new molecular representations for machine learning (ML) based on efficient tight-binding (TB) quantum chemistry ('quantum features') and to connect those representations to various new network architectures. The models will be applied to predict chemically relevant properties of pharmaceutical-type molecules, like conformational and tautomerization energies, pKa values, solubility or partition coefficients. It is a project of a world-wide leading theoretical chemistry group for the development and application of simplified quantum chemical (QC) methods with strong support from the science and technology company Merck with established competence in leveraging extensive chemical data. For computing the quantum features, a new model Hamiltonian (ShellQ) in an extended AO basis set (vDZP) will be developed that is able to reproduce accurately various properties (atomic charge, shell population, bond order, dipole moment, polarizability) from a reference DFT calculation and is still generally applicable to the whole periodic table including organometallic systems. It accounts for the first time in a semiempirical context for fundamental physical effects like orbital contraction and electronic polarization. It is combined with established continuum solvation theories to model solvated molecules. Further main aspects of the proposal are the optimization of the neural network architecture based on ShellQ features, development of feature representation, the automatized generation of molecular training data sets, and state-of-the-art multitask-learning inspired from image recognition algorithms. In general, we follow a Delta-ML strategy where a correction term to a fast QC calculation (typically the established GFN-xTB or GFN-FF methods) based on the available features is computed by the network. This entire approach is supposed to provide efficiency and accuracy for a potentially wide range of chemical properties.