分子相似性是虚拟筛选技术的重要环节,在计算机辅助药物设计中有着关键作用.在2D Fingerprint相似性判别过程中,一些典型的分子相似性评估过程使用了Hash函数进行分子指纹映射,然而Hash函数固有的冲突问题极易降低分子指纹的映射精度.提出了一种基于计数型布隆过滤器的指纹映射方法,对相似性判别过程进行了有效改进,并采用DUD LIB VS 1.0数据集对改进方法进行了比较验证,将ROCE,AUC,awROCE,awAUC值作为评价标准.与其他原始分子相似性方法相比,所提方法有效地提高了相似性判断的准确性和骨架跃迁能力.
Research of Molecular Similarity Algorithm Based on Counting Bloom Filter
Molecular similarity is an important part of the virtual screening technology,and plays a key role in computer-aid drug design.In the process of 2D Fingerprint similarity assessment,some typical molecular similarity assessment use the Hash function in the process of molecular fingerprint mapping.However,the inherent conflict of Hash function easily reduces the precision of molecular fingerprint mapping.In this paper,a fingerprint mapping method based on counting bloom filter was adopted to effectively reduce the probability of fingerprint space mapping conflict and improved the similarity assessment process.To effectively evaluate molecular similarity,the improved method,which uses a tailored version of DUD (DUD LIB VS.1.0 sets),was validated by comparing with experimental results,using ROCE (Receiver Operating Characteristics Enrichment),AUC (Area Under Curve),awROCE and awAUC value as the evaluation standard.Compared with the other original molecular similarity method,the experimental result shows that improved method is still competitive in precision and scaffold hopping potential evaluation standard.
Molecular similarity2D FingerprintsCounting bloom filterVirtual screeningComputer-aid drug design
NETL
NSTL
万方数据
