Enriched biological activity information of compounds in large and free-accessing chemical databases like the PubChem enable the use of target information in 2D similarity search for database screening. The specification of target in the query and the use of bioactivity information about that target to refine the search hold the possibility of enhancing both search recall and precision. This search paradigm also applies to nearest neighbor (NN) searches with multiple bioactive reference structures. A large-scale similarity search investigation has been carried out on 208 well-defined activity classes extracted from PubChem Bioassay Database. The analysis was performed to compare the search performance of three strategies of 2D similarity search approaches, which are non-iterative (single reference) search, iterative NN search with multiple active references, and the neighbor classification (NC) search, which can be regarded as the combination of NN search with active references and reversed NN search with inactive references. The profile concept was also introduced in three strategies of search to examine whether profile based 2D similarity search can further benefit the search performance or lead to computational saving due to a more compact representation of multiple chemical structures. We observed that although NC and profile based NC searches may cause false rejection of active hits, they improve the precision performance in statistical significance, and therefore outperform NN search and profile based NN searches, and non-iterative searches in general. In addition, although the comprehensive search performance of NC are slightly better than that of profile based NC searches, pre-clustering and pre-forming  profiles make profile based NC searches more implementable than fingerprint based NC for advanced similarity search on public chemical database with heavy query burdens. We suggest that NC search and similarly profile based NC search approaches are promising target enhanced similarity search approaches that balance both recall and precision performance.

Using assay information in PubChem to improve 2D similarity searching

248th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.