使用POWGAN的新型分子设计,这是一个政策优化的Wasserstein生成对抗网络
Bruno Macedo1,2, Inês Ribeiro Vaz3,4,5, Tiago Taveira Gomes3,4,6
1Faculty of Medicine, University of Porto, Porto, Portugal. up200601848@up.pt.
Journal of cheminformatics
|December 2, 2025
概括
我们开发了政策优化瓦斯斯坦GAN (POWGAN) 来产生新的,以财产为导向的分子. 这种人工智能方法增强了分子连接性和药物相似性,促进了药物发现的生成化学.
科学领域:
- 化学中的人工智能.
- 生成性对抗性网络 (GANs) 是一个
- 药物发现和开发 药物发现和开发
背景情况:
- 现有的强化导向生成对抗网络 (GAN) 面临着大规模生产非碎片化,属性导向分子的挑战.
- 在优化分子性质和保持生成模型中的样本多样性之间经常存在权衡.
- 目前的GAN很难同时平衡分子连接性,新性和药物相似性.
研究的目的:
- 引入政策优化瓦斯斯坦GAN (POWGAN),这是一个适应性奖励扩展策略,用于改进分子生成.
- 提高生成模型在规模上生产有效,新和属性导向分子的能力.
- 在不损害多样性的情况下克服当前GANs在分子碎片化和属性优化方面的局限性.
主要方法:
- 开发了POWGAN,一个基于图形的生成器,将动态缩放的奖励纳入对抗训练.
- 将POWGAN集成到MedGAN架构中,创建R-MedGAN,以针对图形连接 (非碎片化).
- 评估了基诺林,英多尔和имидазол支架上的模型性能,评估了连接性,新性,独特性和药物相似性 (SAS,LogP).
主要成果:
- R-MedGAN实现了1.00个完全连接的类分子,比基线显著改善 (0.62),同时保持了高新性 (0.93) 和独特性 (0.95).
- 该模型产生了超过12,000个新的类分子,填充了化学空间以前未经探索的区域.
- 药物类似性质大大改善:合成可访问性得分 (SAS) 从8%增加到65%,脂友性 (LogP) 从17%增加到45%.
结论:
- POWGAN和R-MedGAN有效地解决了分子生成GAN的局限性,通过使分子拓和属性的同时优化.
- 适应性奖励扩展策略增强了生成吞吐量,保留了多样性,并改善了药物相似性,在不同的分子支架上展示了可概括性.
- 这项工作提供了一个强大的,可扩展的平台,用于高吞吐量,目标导向的化学探索,推进人工智能驱动药物发现的最先进技术.
相关概念视频
Ziegler–Natta Chain-Growth Polymerization: Overview
3.9K
Ziegler–Natta polymerization is another form of addition or chain‐growth polymerization used for synthesizing linear polymers over branched polymers. The catalyst used for polymerization is the Ziegler–Natta catalyst, named after Karl Ziegler and Giulio Natta, who developed it in 1953. This catalyst is an organometallic complex of titanium tetrachloride and triethyl aluminum, with the active form of the catalyst being an alkyl titanium compound. Using the Ziegler–Natta...
3.9K
Molecular Weight of Step-Growth Polymers
2.7K
Step growth polymerization involves bi or multifunctional monomers. Bifunctional monomers react to form linear step growth polymers, whereas multifunctional monomers react to form non-linear or branched polymers.
As the step-growth polymerization involves step-wise condensation of monomers, the molecular weight also builds up eventually. Consequently, high molecular weight polymers are obtained at the late stages of the polymerization, where 99% of monomers have been consumed.
The extent of the...
As the step-growth polymerization involves step-wise condensation of monomers, the molecular weight also builds up eventually. Consequently, high molecular weight polymers are obtained at the late stages of the polymerization, where 99% of monomers have been consumed.
The extent of the...
2.7K
Radical Chain-Growth Polymerization: Overview
3.1K
Chain-growth or addition polymerization is successive addition reactions of monomers with a polymer chain. In radical chain-growth polymerization, the reaction proceeds via a free-radical intermediate. The free radical is formed from radical initiators, which spontaneously generate free radicals by homolytic fission. Organic peroxides (such as dibenzoyl peroxide, as shown in Figure 1) or azo compounds are popular radical initiators. A low concentration ratio of radical initiator to monomer is...
3.1K
Anionic Chain-Growth Polymerization: Mechanism
2.4K
The mechanism for anionic chain-growth polymerization involves initiation, propagation, and termination steps. In the initiation step, a nucleophilic anion, such as butyl lithium, initiates the polymerization process by attacking the π bond of the vinylic monomer. As a result, a carbanion, stabilized by the electron‐withdrawing group, is generated. The resulting carbanion acts as a Michael donor in the propagation step and attacks the second vinylic monomer, which acts as a Michael...
2.4K
Radical Chain-Growth Polymerization: Mechanism
3.3K
The radical chain-growth polymerization mechanism consists of three steps: initiation, propagation, and termination of polymerization. The polymerization initiates when a free radical generated from the radical initiator adds to the unsaturated bond in the monomer. The unpaired electron of the free radical and one π electron in the unsaturated bond creates a σ bond between the free radical and the monomer. As a result, the other π electron in the unsaturated bond converts this species into...
3.3K
Predicting Molecular Geometry
44.5K
VSEPR Theory for Determination of Electron Pair Geometries
44.5K


