分子机器学习方法对酶选择性C-H键激活反应:从生成AI到实验验证
Ajnabiul Hoque1, Taiwei Chang2, Jin-Quan Yu2
1Department of Chemistry, Indian Institute of Technology Bombay Powai Mumbai 400076 India sunoj@chem.iitb.ac.in.
Chemical science
|June 25, 2025
概括
分子机器学习 (ML) 模型可以预测化学反应,并设计用于不对称催化物的新配体. 实验验证证证实了ML产生的反应与预测一致,突出了ML.
科学领域:
- 计算化学计算化学
- 机器学习 机器学习
- 有机合成 有机合成
背景情况:
- 机器学习 (ML) 越来越多地应用于化学反应预测,但挑战包括小型,稀疏和倾斜的数据集.
- 现有的ML模型擅长预测已知的反应;然而,使用深度生成模型用于新反应发现和验证的探索较少.
- 催化不对称反应,特别是β-C(sp3)-H激活,至关重要但复杂,需要高效的预测和探索工具.
研究的目的:
- 开发和验证深度学习模型,用于预测催化不对称β-C(sp3)-H激活反应的结果.
- 探索使用生成型深度学习模型来设计新奇的性配体,以指导新的反应发现.
- 通过实验验证来评估ML产生的反应预测和配体设计的可靠性和新性.
主要方法:
- 采用转移学习方法,使用在大型分子数据集上预训练的化学语言模型 (CLM).
- 通过对220个实验报告的非对称β-C(sp3) -H激活反应的数据集,对CLM进行了微调.
- 一个集成预测 (EnP) 模型,包括30个微调的CLM,被开发用于预测反应选活性 (% ee).
- 一个单独的CLM在77个已知的性接体上进行了微调,以产生新的接体候选者.
主要成果:
- EnP模型在预测试验组反应的% ee方面表现出高可靠性.
- 产生干的CLM产生了具有高预测有效性的新干.
- 概念验证湿实验室实验验证了大多数ML产生的反应,显示出与EnP预测的良好一致.
结论:
- 深度学习,通过转移学习和集合预测,可以有效地指导发现新的化学反应和连接体.
- 该研究强调了ML在推进非对称催化剂的联体设计方面的潜力.
- 领域专业知识对于在机器学习驱动的反应开发中做出关键决策至关重要,平衡计算预测与专家判断.
相关概念视频
Reduction of Alkenes: Asymmetric Catalytic Hydrogenation
3.4K
Catalytic hydrogenation of alkenes is a transition-metal catalyzed reduction of the double bond using molecular hydrogen to give alkanes. The mode of hydrogen addition follows syn stereochemistry.
The metal catalyst used can be either heterogeneous or homogeneous. When hydrogenation of an alkene generates a chiral center, a pair of enantiomeric products is expected to form. However, an enantiomeric excess of one of the products can be facilitated using an enantioselective reaction or an...
The metal catalyst used can be either heterogeneous or homogeneous. When hydrogenation of an alkene generates a chiral center, a pair of enantiomeric products is expected to form. However, an enantiomeric excess of one of the products can be facilitated using an enantioselective reaction or an...
3.4K
Cycloaddition Reactions: MO Requirements for Thermal Activation
3.7K
Thermal cycloadditions are reactions where the source of activation energy needed to initiate the reaction is provided in the form of heat. A typical example of a thermally-allowed cycloaddition is the Diels–Alder reaction, which is a [4 + 2] cycloaddition. In contrast, a [2 + 2] cycloaddition is thermally forbidden.
3.7K
Conjugate Addition of Enolates: Michael Addition
2.9K
The attack of a nucleophile at the β carbon of an α,β-unsaturated carbonyl compound is called conjugate addition. Conjugate addition reactions of active methylene compounds, such as β-diketones, β-keto esters, β-keto nitriles, and α-nitro ketones, are called Michael addition reactions.
2.9K
Prochirality
4.0K
The concept of prochirality leads to the nomenclature of the individual faces of a molecule and plays a crucial role in the enantioselective reaction. It is a concept where two or more achiral molecules react to produce chiral products. A typical process is the reaction of an achiral ketone to generate a chiral alcohol. Here, the achiral reactant reacts with an achiral reducing agent, sodium borohydride, to generate an equimolar mixture of the chiral enantiomers of the product. For example, an...
4.0K
Aldehydes and Ketones with Amines: Enamine Formation Mechanism
5.9K
Enamine formation involves the addition of carbonyl compounds to a secondary amine through a series of reactions. The mechanism begins with the generation of carbinolamine, a nucleophilic attack followed by several proton transfer reactions. The hydroxyl group of the carbinolamine is converted into water to make a better leaving group that can push the reaction forward by eliminating a water molecule. In enamine formation, the last step involves the abstraction of a proton from the α carbon to...
5.9K
Synthesis of α-Substituted Carbonyl Compounds: The Stork Enamine Reaction
3.5K
α-Substituted ketones or aldehydes can be synthesized from enamines by the Stork enamine reaction, named after its pioneer Gilbert Stork. Enamines are useful synthetic intermediates where the lone pair on nitrogen is in conjugation with the C=C bond. They resemble enolate ions, as the resonance forms of both species have a nucleophilic α carbon.
3.5K


