分子图组装化学空间的快速探索
Rapid Exploration of Assembly Chemical Space of Molecular Graphs
AI总结:
本文提出一种精确可扩展的分子组装指数计算算法,通过最大重复子图优先、状态重用和动态规划分支定界实现比先前方法快六个数量级的加速,并基于此引入联合组装重叠度度量,使分子组装指数成为大规模化学空间探索的实用工具。
摘要:
量化构建分子图的难度对于生物特征检测、化学复杂性和化学信息学具有重要意义。我们提出了一种精确且可扩展的算法来计算分子组装指数(MA),该算法优先考虑最大的重复子图,使用边列表的“组装状态”数组表示碎片化过程,通过哈希/有向无环图重用状态,并利用条件加法链下界引导的动态规划分支定界来剪枝搜索空间。对于分子量大于500 Da的有机分子,我们的方法比先前方法快最多六个数量级,并且能够在以往算法会超时的情况下得到精确的MA值。我们对约30万个COCONUT天然产物(键数小于50)计算了MA直至收敛,分析了时间和内存的扩展规律。最后,我们利用算法的高速度计算联合组装空间,并引入了联合组装重叠度(JAO)——一种类似Jaccard的度量指标,强调全局骨架的重复利用,并证明JAO与基于ECFP指纹和最大公共子结构(MCS)的Tanimoto相似度产生显著不同的排序结果(例如在270-380 Da的类固醇和短肽中),能够捕捉超出局部环境的子结构相似性。综上所述,这些进展使分子组装指数成为大规模探索化学空间的实用工具。
英文摘要:
Quantifying how hard it is to build a molecular graph matters for biosignature detection, chemical complexity, and cheminformatics. We present an exact, scalable algorithm to compute the molecular assembly index (MA) which prioritizes the largest duplicate subgraphs, represents fragmentation with an 'assembly state' array of edge-lists, reuses states via hashing/DAGs, and prunes the search using a dynamic-programming branch-and-bound guided by a conditional-addition-chain lower bound. For organic molecules in the greater than 500 Da range our approach is up to six orders of magnitude faster than prior methods and yields exact MAs where previous algorithms would have timed out. We compute MAs to convergence for ~300k COCONUT natural products with <50 bonds, profiling time and memory scaling. Finally, we exploit the speed of our algorithm to calculate joint assembly spaces and introduce the Joint Assembly Overlap (JAO), a Jaccard-like metric that emphasizes global scaffold reuse and show that the JAO yields substantially different rankings from Tanimoto similarity with ECFP fingerprints and MCS (e.g. in steroids 270-380/Da and short peptides), accounting for substructural similarity beyond local environments. Together, these advances turn the molecular assembly index into a practical tool for large-scale exploration of chemical space.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。