Efficient synthon-based modular screening of Giga-to-Terra-scale virtual libraries

摘要 我们建议的目标是开发一个可扩展的平台，用于基于结构的千兆和兆兆虚拟筛选- 扩大类药物化合物库，使高质量候选药物的发现变得流畅。可用性 蛋白质靶标结构和虚拟化合物(&gt；100亿)位置的千兆级实空间库 基于对接的虚拟筛选是药物发现的关键范例。然而，Giga的计算成本- 规模化筛选成为限制筛选文库进一步发展的主要瓶颈。最近，我们有 推出了一种高度可扩展的基于Synthon的技术V-Synths，它执行分层结构- 基于对可用于合成的(真实)文库的筛选(Sadybekov等人，自然接受)。 通过反复筛选合成子-支架组合，V-Synths方法使快速 在仅执行对接的情况下检测千兆级化学空间中得分最高的化合物 图书馆的一小部分(~200万)。对V-Synths的第一次测试显示，在 计算基准并显著提高大麻素受体CB2和CB2的实验命中率 ROCK1激酶靶点，同时需要的计算资源比标准虚拟筛查少100倍。 在这些初步结果的基础上，我们的建议旨在：(1)进一步开发全自动V- Synths算法，对其参数进行了优化，并将其扩展到Tera规模的真实库中。(2)申请及 在一组不同类别的治疗靶点上实验验证V-Synths方法，这些靶点 包括具有挑战性的靶标，如核苷酸和脂质结合口袋、变构口袋和孤儿 接收器(3)建立算法到开源对接平台的可移植性，以进一步促进V- 辛迪斯在学术实验室中的采用。开源算法将作为Linux的工作流程分发 集群和计算云。该项目的成功完成将使V-Synths成为一家强大的 用于大多数治疗靶点的基于结构的配基发现的计算平台，可扩展到 快速增长的真正的模块化程序库。最重要的是，它将有助于快速进行虚拟筛选 千兆到万亿级的库，整个研究界都可以广泛访问，具有合理的计算能力 资源。

ABSTRACT The goal of our proposal is to develop a scalable platform for structure-based virtual screening of Giga- and Tera- scale drug-like compound libraries, enabling streamlined discovery of high-quality drug candidates. Availability of protein target structures and Giga-scale REAL Space libraries of virtual compounds (>10 billion) position docking-based virtual screening as a key paradigm for drug discovery. However, the computational cost of Giga- scale screening becomes a major bottleneck limiting further growth of the screening libraries. Recently, we have introduced a highly scalable synthon-based technology, V-SYNTHES, which performs hierarchical structure- based screening of REadily AvaiLable for synthesis (REAL) libraries (Sadybekov et al, Nature accepted). By iteratively screening synthon-scaffold combinations, the V-SYNTHES approach makes possible rapid detection of the best-scoring compounds in the Giga-scale chemical space while performing docking of only a small fraction (~2 million) of the library. First tests of V-SYNTHES demonstrated strong enrichment in computational benchmarks and significantly improved experimental hit rates on cannabinoid receptor CB2 and ROCK1 kinase targets, while requiring 100 times less computational resources than standard virtual screenings. Building upon these preliminary results, our proposal aims to: (1) Further develop a fully automated V- SYNTHES algorithm, optimize its parameters and expand it to Tera-scale REAL libraries. (2) Apply and experimentally validate the V-SYNTHES approach on a set of therapeutic targets of different classes, which includes such challenging targets as nucleotide and lipid binding pockets, allosteric pockets, and orphan receptors (3) Establish portability of the algorithm to an open-source docking platform to further facilitate V- SYNTHES adoption in academic labs. The open-source algorithm will be distributed as a workflow for Linux clusters and computing clouds. Successful completion of this project will establish V-SYNTHES as a robust computational platform for structure-based ligand discovery in most classes of therapeutic targets, scaleable for rapidly growing REAL modular libraries. Most importantly, it will help to make fast virtual screening of the Giga-to-Tera-scale libraries broadly accessible for the whole research community with reasonable computational resources.