在高度精选的数据集上,通过基于图形卷积的神经网络预测水溶性
Nadin Ulrich1,2, Karsten Voigt3, Anton Kudria4
1Department of Exposure Science, Helmholtz Centre for Environmental Research-UFZ, Permoserstrasse 15, 04318, Leipzig, Germany. nadin.ulrich@ufz.de.
Journal of cheminformatics
|April 21, 2025
概括
一个新的图形卷积神经网络 (GNN) 使用精准的数据集准确预测化学水溶性 (log Sw). 该工具有助于填补环境化学和药物设计应用的数据缺口.
科学领域:
- 环境化学环境化学
- 毒理学 毒理学 毒理学
- 药物设计 药物设计
背景情况:
- 水溶性是环境化学,毒理学和药物设计的关键物理化学性质.
- 现有的水溶性实验数据是不完整的,需要更广泛的化学空间覆盖的预测工具.
- 准确预测水溶性对于填补化学数据库数据缺口至关重要.
研究的目的:
- 使用图形卷积神经网络 (GNN) 开发一个强大的水溶性 (log Sw) 预测模型.
- 通过结合和清理现有数据库和文献数据,创建一个高度精选的数据集.
- 将GNN模型的性能与实验数据进行评估,并与现有的预测工具进行比较.
主要方法:
- 一个图形卷积神经网络 (GNN) 模型被开发用于预测水溶性 (log Sw).
- 从AqSolDB和文献中策划了一个全面的数据集,分别包括7605个和2195个数据点.
- 使用培训 (70%),验证 (20%) 和测试 (10%) 集的五重交叉验证策略构建了一个共识GNN.
主要成果:
- 在一个独立的测试组中,共识GNN实现了高预测性能,R2 = 0.901,Q2 = 0.896,RMSE = 0.657.
- 该模型的准确性可与实验误差范围为0.5到0.6日志单位相提并论.
- 该研究提供了对模型的应用领域和性能相对于其他预测工具的洞察.
结论:
- 开发的GNN模型提供了一种可靠的方法,用于预测广泛的化学品的水溶性.
- 数据整理过程中发现并纠正了实验水溶性数据中的错误.
- 这种预测工具可以通过解决数据稀缺问题,显著帮助环境化学,毒理学和药物设计的研究人员.
相关概念视频
Solubility
17.1K
Solution, Solubility, and Solubility Equilibrium
A solution is a homogeneous mixture composed of a solvent, the major component, and a solute, the minor component. The physical state of a solution—solid, liquid, or gas—is typically the same as that of the solvent. Solute concentrations are often described with qualitative terms such as dilute (of relatively low concentration) and concentrated (of relatively high concentration).
In a solution, the solute particles (molecules,...
A solution is a homogeneous mixture composed of a solvent, the major component, and a solute, the minor component. The physical state of a solution—solid, liquid, or gas—is typically the same as that of the solvent. Solute concentrations are often described with qualitative terms such as dilute (of relatively low concentration) and concentrated (of relatively high concentration).
In a solution, the solute particles (molecules,...
17.1K
Solution Formation
30.8K
There is no one solvent that can dissolve every type of solute. Some substances that readily dissolve in a certain solvent might be insoluble in a different solvent. A simple way to predict which substances dissolve in which solvent is the phrase "like dissolves like". This means that polar substances, such as salt and sugar, dissolve in a polar substance like water. In contrast, non-polar substances are more soluble in non-polar solvents such as carbon tetrachloride.
This selective...
This selective...
30.8K
Solvents
63.8K
A solvent is a substance, most often a liquid, that can dissolve other substances. Here, the substance being dissolved is called a solute. When a solvent and a solute combine, they form a solution - a homogenous mixture of both the solvent and the solute. Water is a universal biological solvent. Its polar structure allows it to dissolve many other polar compounds. The ability of water to dissolve is governed by a balance between water molecules binding to each other and binding to the solute.
A...
A...
63.8K
Water and Mineral Acquisition
28.7K
Specialized tissues in plant roots have evolved to capture water, minerals, and some ions from the soil. Roots exhibit a variety of branching patterns that facilitate this process. The outermost root cells have specialized structures called root hairs that increase the root surface, thus increasing soil contact. Water can passively cross into roots, as the concentration of water in the soil is higher than that of the root tissue. Minerals, in contrast, are actively transported into root cells.
28.7K
Factors Affecting Solubility
32.8K
Compared with pure water, the solubility of an ionic compound is less in aqueous solutions containing a common ion (one also produced by dissolution of the ionic compound). This is an example of a phenomenon known as the common ion effect, which is a consequence of the law of mass action that may be explained using Le Chȃtelier’s principle. Consider the dissolution of silver iodide:
32.8K
Solubility Equilibria: Ionic Product of Water
904
Pure water is a weak electrolyte; only a small amount ionizes into hydrogen and hydroxide ions. At any given temperature, the concentration of undissociated water is almost constant, so the ionic product of water is the product of the hydrogen and hydroxide ion concentrations, denoted as Kw. The square root of Kw gives the individual ion concentrations.
The ionic product of water varies with temperature, and its value is 1.0 x 10−14 at standard experimental conditions. Per Le...
The ionic product of water varies with temperature, and its value is 1.0 x 10−14 at standard experimental conditions. Per Le...
904


