有机分子的三维自回归生成机器学习中的结构偏差
Zsuzsanna Koczor-Benda1, Joe Gilkes1,2, Francesco Bartucca1
1Department of Chemistry, University of Warwick, Coventry CV4 7AL, U.K.
Journal of chemical information and modeling
|June 25, 2025
概括
像G-SchNet这样的生成机器学习模型可以设计新型分子,但它们经常显示偏差. 这项研究揭示了G-SchNet产生具有较少和键和更多异原子的分子,影响化学空间和特性.
科学领域:
- 计算化学是一种计算化学.
- 材料科学是一种材料科学.
- 机器学习是机器学习.
背景情况:
- 生成型机器学习模型越来越多地用于设计新型分子和材料.
- 三维结构生成对于量子化学工作流程和属性预测至关重要.
- 模型性能评估传统上侧重于新性,有效性和独特性,但化学空间复制也至关重要.
研究的目的:
- 调查G-SchNet自回归模型能够复制训练数据的化学空间和属性分布的能力.
- 评估G-SchNet的生成偏差对分子性质和化学空间覆盖的影响.
- 探索减轻生成偏差的方法,例如功能组约束和复合数据集.
主要方法:
- 对元素组成,大小,键长,功能组和化学空间分布进行分析,用于训练和生成分子.
- 化学空间的主要组件分析 (PCA) 以确定生成偏差.
- 决策树模型的应用,以区分培训和生成的数据,并揭示化学差异.
主要成果:
- G-SchNet 呈现出生成偏差,产生和度较低且含有更多异构原子的分子,无论超参数或训练数据分布如何.
- 在生成的集合中,纯粹的异质分子在很大程度上缺席.
- 决策树模型成功地确定了生成偏差,并突出了影响电子属性的关键化学差异,如HOMO-LUMO差距.
结论:
- G-SchNet模型在生成分子方面表现出显著的偏差,导致训练数据的化学空间和属性的不准确表示.
- 功能组约束和复合数据集可以部分缓解观察到的生成偏差.
- 了解和解决生成偏差对于使用生成模型准确设计具有所需电子性质的功能分子至关重要.
相关概念视频
Fischer Projections
13.9K
Learning to draw Fischer projections of molecules and understanding their relevance plays a crucial role in the visual depiction of organic molecules. A Fischer projection is a two-dimensional projection on a planar surface to simplify the three-dimensional wedge–dash representation of molecules. This is especially helpful in the case of molecules with multiple chiral centers that can be difficult to draw. Here, all the bonds of interest are represented as horizontal or vertical lines.
13.9K
Predicting Molecular Geometry
36.2K
VSEPR Theory for Determination of Electron Pair Geometries
36.2K
Molecular Models
40.8K
Physical models representing molecular architectures of chemical compounds play essential roles in understanding chemistry. The use of molecular models makes it easier to visualize the structures and shapes of atoms and molecules.
40.8K
Polymer Classification: Stereospecificity
2.7K
Polymerization generates chiral centers along the entire backbone of a polymer chain. Accordingly, the stereochemistry of the substituent group has a significant effect on polymer properties. Polymers formed from monosubstituted alkene monomers feature chiral carbons at every alternate position in the polymer backbone. Relative to the predominant orientation of substituents at the adjacent chiral carbons, the polymer can exist in three different configurations: isotactic, syndiotactic, and...
2.7K
Molecules with Multiple Chiral Centers
12.5K
Molecules that possess multiple chiral centers can afford a large number of stereoisomers. For instance, while some molecules like 2-butanol have one chiral center, defined as a tetrahedral carbon atom with four different substituents attached, several molecules like butane-2,3-diol have multiple chiral centers. A simple formula to predict the number of stereoisomers possible for a molecule with n chiral centers is 2n. However, there can be a lower number where some of the stereoisomers are...
12.5K
Newman Projections
17.8K
Different notations are used to represent the three-dimensional structure of molecules on two-dimensional surfaces. One of the most commonly used representations is the dash-wedge formula. The dashed wedges, solid wedges, and the plane lines indicate the groups situated behind the plane, coming out of the plane, and in the plane, respectively.
The organic molecules rotate across the single bonds leading to numerous temporary three-dimensional structures of varying energy known as...
The organic molecules rotate across the single bonds leading to numerous temporary three-dimensional structures of varying energy known as...
17.8K


