在机器学习中优先考虑数据质量用于热物理性质预测:关于有机化合物的正常沸点的案例研究
Frank T Mtetwa1, Neil F Giles1, W Vincent Wilding1
1Department of Chemical Engineering, Brigham Young University, Provo, Utah 84604, United States.
ACS omega
|November 3, 2025
概括
高质量的数据对于准确的机器学习 (ML) 对正常沸点 (NBP) 的预测至关重要. 严格策划的实验数据为化学过程设计提供了更好的ML模型,而不是更大的,未经策划的数据集.
科学领域:
- 化学工程是化学工程的重要组成部分.
- 计算化学计算化学
- 机器学习 机器学习
背景情况:
- 准确的正常沸点 (NBP) 预测对于化学过程设计和优化至关重要.
- 传统的方法,如集团贡献 (GC) 和定量结构与财产关系 (QSPR),在灵活性和准确性方面存在局限性.
- 机器学习 (ML) 是有前途的,但在训练集之外的组合体的易用性和可靠性方面面临挑战,通常是由于数据质量问题.
研究的目的:
- 调查数据质量对ML模型性能对NBP预测的影响.
- 为了比较在精选与未精选数据集上训练的ML模型.
- 为可访问的NBP预测开发一个用户友好的工具.
主要方法:
- 训练有素的ML模型使用LightGBM和PyTorch.
- 评估了四种分子特征化方法:RDKit描述符,Joback组和两个图形神经网络 (GNN) 表示.
- 在一个精选的实验数据集 (DIPPR 801) 和一个更大的,未精选的数据集上使用10倍交叉验证来评估模型性能.
- 开发了一个网页应用程序,以便轻松访问表现最佳的模型.
主要成果:
- 在较小,严格策划的数据集上训练的ML模型始终优于在较大,未经策划的数据集上训练的模型.
- 精心策划的数据集方法展示了卓越的准确性,减少偏差和更好的概括能力.
- 数据质量,而不仅仅是数量,被确定为可靠的ML驱动化学性质预测的关键因素.
结论:
- 以数据为中心的方法,优先考虑数据质量,对于开发化学性质预测中的强大的ML模型至关重要.
- 这些发现倡导使用严格策划的实验数据来训练ML模型.
- 提供了一个实用的网络应用程序,使非专家能够使用简单的SMILES公式预测NBP.
相关概念视频
Intermolecular Forces and Physical Properties
26.4K
26.4K
Comparing Intermolecular Forces: Melting Point, Boiling Point, and Miscibility
50.2K
Intermolecular forces are attractive forces that exist between molecules. They dictate several bulk properties, such as melting points, boiling points, and solubilities (miscibilities) of substances. Molar mass, molecular shape, and polarity affect the strength of different intermolecular forces, which influence the magnitude of physical properties across a family of molecules.
Temporary attractive forces like dispersion are present in all molecules, whether they are polar or nonpolar. They...
Temporary attractive forces like dispersion are present in all molecules, whether they are polar or nonpolar. They...
50.2K
Physical Properties of Carboxylic Acid Derivatives
3.2K
Intermolecular forces dictate several physical properties such as boiling points, melting points, solubilities, and so forth. They are classified into four types: ionic forces, hydrogen bonds, dipole–dipole forces, and dispersion forces. Ionic forces are the strongest, while dispersion forces are the weakest.
Among the carboxylic acid derivatives, the boiling points of acid chlorides and esters are very similar and are the lowest in the series. Acid anhydrides have slightly higher boiling...
Among the carboxylic acid derivatives, the boiling points of acid chlorides and esters are very similar and are the lowest in the series. Acid anhydrides have slightly higher boiling...
3.2K
Distillation: Vapor–Liquid Equilibria
4.3K
Distillation is a separation technique that takes advantage of the boiling point properties of disparate elements in a mixture. To perform distillation, we begin by heating a miscible mixture of two liquids with a significant difference in boiling points (at least 20°C). As the solution heats up and reaches the bubble point of the more volatile component, some molecules of the more volatile component transition into the gas phase and travel upward into the condenser, which is a glass tube...
4.3K
Vapor Pressure
38.8K
When a liquid vaporizes in a closed container, gas molecules cannot escape. As these gas phase molecules move randomly about, they will occasionally collide with the surface of the condensed phase, and in some cases, these collisions will result in the molecules re-entering the condensed phase. The change from the gas phase to the liquid is called condensation. When the rate of condensation becomes equal to the rate of vaporization, neither the amount of the liquid nor the amount of the vapor...
38.8K
Physical Properties of Ethers
8.3K
Overview
An ether molecule has a net dipole moment due to the polarity of C–O bonds. Subsequently, boiling points of ethers are lower than those of alcohols of comparable molecular weight and slightly higher than those of hydrocarbons of comparable molecular weight (Table 1).
Ethers can act as hydrogen bond acceptors, making them more water-soluble than hydrocarbons, but since ethers cannot act as hydrogen bond donors, they are much less soluble in water than alcohols. Ethers are considered...
An ether molecule has a net dipole moment due to the polarity of C–O bonds. Subsequently, boiling points of ethers are lower than those of alcohols of comparable molecular weight and slightly higher than those of hydrocarbons of comparable molecular weight (Table 1).
Ethers can act as hydrogen bond acceptors, making them more water-soluble than hydrocarbons, but since ethers cannot act as hydrogen bond donors, they are much less soluble in water than alcohols. Ethers are considered...
8.3K


