After decades of development and nearly continuous use (and abuse) in support of drug discovery efforts, QSAR and related descriptor-based statistical learning methods have earned mixed reviews throughout their checkered past.  Why is this?  In practice, it is necessary for users to choose from a long list of available descriptors and then select a machine learning method with the hope that the resulting model will have a chance of representing the physical effects that actually control the endpoint of interest.   All too often, the resulting models might have been over-trained using small datasets, or could have limited applicability domains.  Inappropriate descriptor choices can also doom such efforts.  Why and when does this happen?  In this talk, the evolution of �Best Practices� in predictive cheminformatics will be illustrated by way of a series of example scenarios, concluding with some guidelines and recommendations for both model builders and end users.

Developing “Best Practices” in predictive cheminformatics for drug-discovery applications

243rd National Meeting (2012)

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