论文部分内容阅读
化合物的子结构检索在计算机辅助药物设计、波普学、化学数据库等领域是不可或缺的工具。然而由于子结构检索是一个NP完备性的问题,获得用户可接受的平均检索时间一直是研究人员十分关注的问题,其方法主要有改进算法和提升硬件条件2个方面。当化学结构数据库的规模达到百万乃至千万级别时,尽管改进算法的方式能够获得一定的检索效率提升,但其提升的空间有限,因而,集群并行方式是大规模化合物子结构检索应用的必然选择。本文以ChemDB Portal的化学子结构检索系统为基础,实现了基于集群并行的化学子结构检索系统,并进行了任务均分、多线程并行等优化。在包含800万个化合物结构的化学结构数据库中,利用5个节点的小型集群,选取10个较为典型的提问结构进行子结构检索测试。测试结果为基于集群的化学子结构检索的平均检索时间由初始单节点时的34.1 min降低为2.75 min,检索效率平均提高12.4倍,表明在大规模乃至超大规模的数据条件下,集群并行化方式能够显著地提高子结构检索系统的执行效率。
Substructure search of compounds is an indispensable tool in the field of computer-aided drug design, popularization, chemical databases and so on. However, because substructure retrieval is an NP completeness problem, it is always a very important issue for researchers to get the average retrieval time acceptable to users. The main methods are improving algorithms and improving hardware conditions. When the scale of the chemical structure database reaches millions or even tens of millions of levels, although the improved algorithm can obtain some retrieval efficiency improvement, the space for its promotion is limited. Therefore, the parallel cluster approach is inevitable for the retrieval of large-scale compound substructures select. Based on ChemDB Portal’s chemical sub-structure retrieval system, this paper realizes the chemical sub-structure retrieval system based on cluster parallelism, and optimizes the task sharing and multithreading parallelism. In a chemical structure database containing 8 million compound structures, a small cluster of 5 nodes was used to select 10 typical query structures for substructure search. The test results showed that the average search time of cluster-based chemical sub-structure retrieval decreased from 34.1 min in the initial single node to 2.75 min, and the retrieval efficiency increased by 12.4 times on average, indicating that the clustering parallelization method under large-scale and even ultra- Can significantly improve the execution efficiency of the substructure search system.