Wendy Warr Interviews Alex Tropsha

Alex Tropsha is Professor and Chair of the Division of Medicinal Chemistry and Natural Products at the University of North Carolina (UNC) School of Pharmacy. He moved to UNC in 1989, having completed an MS in chemistry, a PhD in biochemistry/pharmacology, and post-doctoral work in QSAR and drug design at Moscow State University. He lists his research interests as development of new methodologies and software tools for computer-assisted drug design; and development of a new approach to protein 3D structure analysis and prediction based on the principles of statistical geometry.

Interview

Wendy A. Warr: Our readers probably think of you as a QSAR expert but you don’t mention QSAR in your list of research interests. Why is that?

Alex Tropsha: QSAR has been indeed my major area of research and I am completely dedicated to this subject; however, I have been thinking recently about the difference between, say, a research tool or a field, and a scientific discipline. I think of QSAR as an area of research and a tool, but it does not amount to a discipline. This tool (albeit very complex) embraces concepts from many disciplines (such as statistics, chemistry, pharmacology) and in my mind is one of several major tools used in a broader area of computational drug design, or generally speaking, drug design, which I do regard as my major area of research interest. So, in brief, I distinguish a goal (drug discovery) from the means (QSAR modeling).

WAW: We are all aware of your article “Beware of q2”. What is big in QSAR right now?

AT: That paper identified one of the major deficiencies of QSAR modeling, i.e., lack of external predictivity of models based solely on the accuracy of fit to experimental data. This problem of identifying models that do have confirmed external predictive power remains the biggest challenge of QSAR modeling. In my opinion, without concurrently addressing this challenge and proving, by means of experimental validation, that QSAR predictions are accurate, all other typical areas of QSAR investigations (descriptors, data mining approaches, and especially, model interpretation) are quite meaningless. Due to rapidly growing availability of very large and diverse data sets or chemical libraries tested in multiple biological assays, the scientific problems that the field of QSAR is facing are quite complex. The big technical issues relate to methodologies for model validation, model applicability domain, multi-objective model optimization, outlier detection handling, and the use of QSAR models iteratively as hypothesis generation/virtual screening/compound prioritization, decision support tools for experimental biomolecular sciences