Special Issue on Reaction Informatics and Chemical Space.
Special Issue on Reaction Informatics and Chemical Space.
复制标题
反应信息学和化学空间特刊。
DOI：
10.1021/acs.jcim.2c00390
复制
发表时间：
2022
影响因子：
5.6
通讯作者：
Warr,Wendy
中科院分区：
文献类型：
--
作者：
Rarey,Matthias;Nicklaus,MarcC;Warr,Wendy
化学信息学家生活在激动人心的时代，正如所有从事自然科学和计算机科学交叉领域工作的科学家一样。数字数据收集的规模和数量不断增加，包括机器学习在内的新颖数据分析开辟了收集这些数据的途径。在过去的十年中，使这种质变成为可能的要素汇集在一起：更多的数据可用性和更好的（公平）标准； GPU 和云计算形式的硬件取得了惊人的进步；改进的算法；机器学习领域取得显着突破。大约一年前，我们开始收集有关反应信息学和化学空间的特刊。美国国立卫生研究院 (NIH) 举办的两场数字研讨会显示了全世界对这些化学信息学领域的兴趣。这些研讨会是由最近在计算机辅助合成预测方面取得的成功、有机化学生成模型的发明以及大型组合化学空间（包括处理它们的工具）的可用性引发的。本期特刊中包含的论文只能及时提供一个快照，突出显示化学信息学中一些令人兴奋的发展。有几篇论文涉及应用现代机器学习技术生成新的化学结构。这些方法的优点是生成过程可以偏向于所需的属性。 Kaitoh 和 Yamanishi 1 在结构生成过程中将注意力集中在分子支架上。在同一方向上，Fialkova 等人的 Lib-INVENT 方法。 2 支持机器学习驱动的库设计。 Bagal 等人将 Transformer 模型作为机器学习中的一项最新技术应用于分子生成。 3 并且还探索了新的算法，如 Mercado 等人的论文中所述。 4 能够处理包括反应在内的大型化合物集合的软件系统是化学空间计算以及为其生成而开发的系统的关键要素。 CGRdb2。 0 是一个用于化学数据存储的经典数据库系统，包括由 Gimadiev 等人开发的反应。 5 本期特刊介绍了两个用于生成化学碎片空间的开源软件系统，即 Wahl 和 Sander 6 的 OpenChemLib 系统以及 Varnek 小组开发的 SynthI。 7 Schmidt 等人首次提出了一种在大型化学片段空间中搜索最大公共子结构的精确算法。 8 作为本期特刊的副主编和客座编辑，我们的印象是，关于超大规模数据收集和化学空间的科学非常重要。因此，我们与 C. Nicolaou 一起决定撰写一篇有关该领域的评论文章。 9
Cheminformaticians live in exciting times as all scientist do who are working at the interface of natural and computer sciences. Digital data collections increase in size and numbers and novel data analytics including machine learning opens a route for harvesting this data. The ingredients to make this qualitative change possible came together during the past decade: more data availability with better (FAIR) standards; astonishing hardware advances in the form of GPU and cloud computing; improved algorithms; remarkable breakthroughs in machine learning. About a year ago we started our endeavor collecting a special issue on Reaction Informatics and Chemical Space. Two digital workshops at the NIH showed the worldwide interest in these areas of cheminformatics. The workshops were triggered by recent successes in computeraided synthesis predictions, the invention of generative models in organic chemistry, and the availability of large combinatorial chemical spaces including the tools to handle them. The papers included in this special issue can only give a snapshot in time highlighting some of the exciting developments in cheminformatics. Several papers deal with the generation of new chemical structures applying modern machine learning techniques. The advantage of these approaches are that the generation process can be biased toward desired properties. Kaitoh and Yamanishi 1 keep their focus on the molecular scaffold during structure generation. In the same direction, the Lib-INVENT approach by Fialkova et al. 2 enables machine-learning driven library design. Transformer models as a recent technique in machine learning are applied to molecule generation by Bagal et al. 3 and also new algorithms are explored, as in the paper by Mercado et al. 4 Software systems able to handle large compound collections including reactions are a key element to computations in chemical space as well as systems developed for its generation. CGRdb2. 0 is a classical database system for chemical data storage including reactions developed by Gimadiev et al. 5 Two open-source software systems to generate chemical fragment spaces are presented in this special issue, the OpenChemLib system by Wahl and Sander 6 as well SynthI developed in the Varnek group. 7 For the first time, a precise algorithm for maximum common substructure searching in large chemical fragment spaces is presented by Schmidt et al. 8 We, the associate editor and guest editors of this special issue, got the impression that the science about ultralarge data collections and chemical spaces is very vital. We, together with C. Nicolaou, therefore decided to contribute a Review article about this field. 9