Park, Keunwan Ko, Young-Joon Durai, Prasannavenkatesh Pan, Cheol-Ho 2024-01-19T18:34:30Z 2024-01-19T18:34:30Z 2021-08-31 2019-11-18 0305-1048 https://pubs.kist.re.kr/handle/201004/119317 Chemical similarity searching is a basic research tool that can be used to find small molecules which are similar in shape to known active molecules. Despite its popularity, the retrieval of local molecular features that are critical to functional activity related to target binding often fails. To overcome this limitation, we developed a novel machine learning-based chemical binding similarity score by using various evolutionary relationships of binding targets. The chemical similarity was defined by the probability of chemical compounds binding to identical targets. Comprehensive and heterogeneous multiple target-binding chemical data were integrated into a paired data format and processed using multiple classification similarity-learning models with various levels of target evolutionary information. Encoding evolutionary information to chemical compounds through their binding targets substantially expanded available chemical-target interaction data and significantly improved model performance. The output probability of our integrated model, referred to as ensemble evolutionary chemical binding similarity (ensECBS), was effective for finding hidden chemical relationships. The developed method can serve as a novel chemical similarity tool that uses evolutionarily conserved target binding information. English OXFORD UNIV PRESS SEARCH DATABASE REPRESENTATIONS PERFORMANCE FRAMEWORK 2D Machine learning-based chemical binding similarity using evolutionary relationships of target genes Article 10.1093/nar/gkz743 1 NUCLEIC ACIDS RESEARCH, v.47, no.20 NUCLEIC ACIDS RESEARCH 47 20 scie scopus 000501735500006 2-s2.0-85074873796 Biochemistry & Molecular Biology Biochemistry & Molecular Biology Article SEARCH DATABASE REPRESENTATIONS PERFORMANCE FRAMEWORK 2D