General selectivity trends across targets have implications to  target selection, compound prioritization, understanding toxicity and patient tailoring. Large body of research focuses on understanding compound polypharmacology and general implications of compound derived target spaces.  We highlight the difficulty of drawing general conclusions from compound data by highlighting differences in interpretation of data derived from different profiling panels.  We demonstrated that general conclusions about target space or general inhibitor class promiscuity are both profiling data and inhibitor set dependent using kinase set.  We also introduce new metrics for comparing selectivity predictions and use them to contrast different profiling data to predictive models. Our results also indicate that computational models generated from profiling can, in some cases, be used in lieu of having literature/other panel data on a set of molecules. We also show that computational models built from one data set cannot be reliably used to inform on another data set.  In particular, utilizing literature or publically available data sets to build computational models is bound to carry large discrepancy from the intended use experiment.  At the same time, while understanding compound data derived target space characterization is data set depended, we demonstrate value generated from visualization and mining of target spaces for a given compound collection.  We show the methodology that allows to visualize Chembl and internal target space that allows for target combination hypothesis generation.

What can we learn from profiling and predictive data � implications to target space and selectivity assessment

246th National Meeting (2013)

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