Mining Chemical Reactions Using Neighborhood Behavior And Condensed Reactions Graphs Approachesстатья
Статья опубликована в высокорейтинговом журнале
Информация о цитировании статьи получена из
Web of Science,
Scopus
Статья опубликована в журнале из списка Web of Science и/или Scopus
Дата последнего поиска статьи во внешних источниках: 3 декабря 2017 г.
Аннотация:This work addresses the problem of similarity search and classification of
chemical reactions using Neighborhood Behavior (NB) and Condensed Graphs of
Reaction (CGR) approaches. The CGR formalism represents chemical reactions as a
classical molecular graph with dynamic bonds, enabling descriptor calculations on this
graph. Different types of the ISIDA fragment descriptors generated for CGRs in
combination with two metrics − Tanimoto and Euclidean − were considered as chemical
spaces, to serve for reaction dissimilarity scoring. The NB method has been used to select an optimal combination of descriptors
which distinguish different types of chemical reactions in a database containing 8544 reactions of 9 classes. Relevance of NB
analysis has been validated in generic (multiclass) similarity search and in clustering with Self-Organizing Maps (SOM). NB -
compliant sets of descriptors were shown to display enhanced mapping propensities, allowing the construction of better Self-
Organizing Maps and similarity searches (NB and classical similarity search criteria − AUC ROC − correlate at a level of 0.7).
The analysis of the SOM clusters proved chemically meaningful CGR substructures representing specific reaction signatures.