In this webinar, Dragos will introduce Chemical Library Spaces (CLS), featuring entire libraries as stand-alone objects; CLS are to support the management of a portfolio of libraries — just like Chemical Spaces are used to manage a portfolio of drug-like molecules. CLS rely on vectorial library representations obtained from Generative Topographic Mapping (GTM): GTMs are basically fuzzy-logics-endowed analogs of Kohonen maps, where each item (molecule) is partially associated to several nodes according to a real vector of “responsibilities” summing up to 1.0. With Neighborhood Behavior-compliant GTMs, similar molecules will display similar responsibility vectors and map similarly on the GTM. Therefore, cumulative responsibility vectors of chemical library members are excellent descriptors of libraries: They encode the major chemical space zones, associated to specific “chemotypes” covered by each library. They can be further modulated by weighing according to experimental or calculated molecular properties, making this a flexible and intuitive tool for library analysis in a “tunable” CLS that can be chosen to specifically address the medicinal chemists’ problems.
CLS may itself be “cartographed”. Thousands of DNA-encoded libraries (DELs), for example, can be “located” alongside many other commercial, public or in-house compound libraries, on µGTM landscapes, constructed by dimensionality reduction of CLS vectors (µGTM is used to highlight that the input CLS vectors are themselves originating from standard chemical space GTMs). µGTM landscapes are colored by library-specific properties, seamlessly combining similarity and property characteristics in a same view, and herewith facilitating the selection of best project-specific libraries.
Speaker: Dr. Dragos Horvath, University of Strasbourg, France
