域信息图神经网络:一个量子化学案例研究
Jay Paul Morgan1, Adeline Paiement1, Christian Klinke2
1Université de Toulon, Aix Marseille Univ, CNRS, LIS, Marseille, France; Department of Computer Science, Swansea University, Swansea, SA2 8PP, United Kingdom.
概括
我们将领域知识集成到图形神经网络 (GNN) 中,用于化学系统能量预测. 这种方法通过结合关系类型和物理量来提高准确性和概括性,改进机器学习模型.
科学领域:
- 计算化学的计算化学
- 机器学习 机器学习
- 材料科学 材料科学 材料科学
背景情况:
- 图形神经网络 (GNN) 是分析复杂系统的强大工具.
- 将域名知识集成到GNN中可以提高它们的准确性和概括能力.
- 估计化学系统中的潜在能量对于理解分子和晶体行为至关重要.
研究的目的:
- 探索将先前的领域知识整合到GNN设计中的策略.
- 为了提高化学系统能量估计的GNN的准确性和概括性.
- 调查将关系类型和物理量纳入GNN的影响.
主要方法:
- 开发了GNN,利用对节点之间的不同关系类型 (例如化学键) 的知识.
- 基于关系类型实施专用消息生成和内部状态更新策略.
- 利用多任务学习 (MTL) 来限制学习特征的物理相关性.
- 在三个不同的GNN架构上测试了该方法,并发布了新的数据集.
主要成果:
- 证明集成领域知识显著提高了GNN精度和能源预测的概括性.
- 展示了专门消息制作和状态更新策略的有效性.
- 验证了MTL对增强学习特征的物理相关性的好处.
- 证实了在不同的GNN架构中提出的知识整合方法的普遍适用性.
结论:
- 整合领域知识,特别是关系类型和物理量,是提高GNN在科学应用中的性能的一种可行的策略.
- 提出的方法为开发更准确,更有物理意义的GNN用于化学系统分析提供了强大的框架.
- 代码和数据集的公开发布有助于进一步研究和开发以知识为导向的GNN.
相关概念视频
01:27Inductive Effects on Chemical Shift: Overview
1.2K
The protons in unsubstituted alkanes are strongly shielded with chemical shifts below 1.8 ppm. Methine, methylene, and methyl protons appear at approximately 1.7, 1.2 and 0.7 ppm, while the proton signal from methane appears at 0.23 ppm. An electronegative substituent, such as chlorine, withdraws the electron density from the protons, increasing their chemical shift. Progressive substitution of the hydrogens in methane by chlorine shifts the proton signals increasingly downfield, to 3.05 ppm in...
1.2K
02:27Predicting Molecular Geometry
34.5K
VSEPR Theory for Determination of Electron Pair Geometries
34.5K
02:38Valence Bond Theory and Hybridized Orbitals
19.5K
According to valence bond theory, a covalent bond results when: (1) an orbital on one atom overlaps an orbital on a second atom, and (2) the single electrons in each orbital combine to form an electron pair. The strength of a covalent bond depends on the extent of overlap of the orbitals involved. Maximum overlap is possible when the orbitals overlap on a direct line between the two nuclei.
A σ bond (single bond in a Lewis structure) is a covalent bond in which the electron density is...
A σ bond (single bond in a Lewis structure) is a covalent bond in which the electron density is...
19.5K
01:13Graphing the Wave Function
1.9K
Consider the wave equation for a sinusoidal wave moving in the positive x-direction. The wave equation is a function of both position and time. From the wave equation, two different graphs can be plotted.
1.9K
02:43Quantum Numbers
34.9K
It is said that the energy of an electron in an atom is quantized; that is, it can be equal only to certain specific values and can jump from one energy level to another but not transition smoothly or stay between these levels.
34.9K
02:18Network Covalent Solids
13.5K
Network covalent solids contain a three-dimensional network of covalently bonded atoms as found in the crystal structures of nonmetals like diamond, graphite, silicon, and some covalent compounds, such as silicon dioxide (sand) and silicon carbide (carborundum, the abrasive on sandpaper). Many minerals have networks of covalent bonds.
To break or to melt a covalent network solid, covalent bonds must be broken. Because covalent bonds are relatively strong, covalent network solids are typically...
To break or to melt a covalent network solid, covalent bonds must be broken. Because covalent bonds are relatively strong, covalent network solids are typically...
13.5K