Drug discovery is increasingly influenced by the availability of very large databases of drug-like molecules, comprising millions of compounds from commercial or from theoretical sources awaiting biological evaluation, as well as large collections of bioactive compounds with annotated activities. Working with such large databases requires efficient tools to browse through very large lists of molecules, in particular to rapidly identify structurally similar molecules. Recently we reported a multi-fingerprint browser for the ZINC database allowing to rapidly identify analogs of any screening hit and perform clustering to compose focused sets for hit confirmation.
Here we report the extension of this multi-fingerprint approach to the problem of target prediction searches, i.e. how to find out if a newly identified bioactive molecule or any compound is closely related to molecules with documented bioactivity and therefore likely to interact with the corresponding biological target. Our Polypharmacology browser searches through 4613 groups of at least 10 bioactive molecules with documented activity against a biological target, as listed in ChEMBL, to identify analogs of any query molecule using six different fingerprints and four fingerprint combination (HyperSpace), and displays results groups by targets as lists of bioactive compounds, which allows one to directly estimate whether the identified similarity is meaningful in the examined context.
ChEMBL is an open-access database collecting information on bioactive compounds reported in literatures. As of now it contains ~1.6 millions bioactive compounds reported against more than 10'000 target proteins. For download and more information please visit https://www.ebi.ac.uk/chembl/
Group of Prof. Jean-Louis Reymond
Dept. of Chemistry and Biochemistry