Drug discovery remains one of the most effective mechanisms by which science can have direct impact on public health. In the last 30 years there has been ever increasing interest in the application of computation to this endeavor- ranging from the all-atom simulation of molecular systems to the canonicalization and organization of molecular information. Some approaches, in particular molecular informatics, are now cornerstones of any drug discovery process and are even becoming of regulatory importance. Others are still evolving as techniques and scientific avenues of research. An ever-important aspect of such research is the evaluation of progress - is one method actually better than another? The difficulty of the field of molecular modeling as applied to drug discovery (Computer-Aided Drug Design, or CADD) is that the process is exceedingly long and very expensive. As such, it is usually infeasible to rigorously test one method relative to another. No one is going to run, for example, double blind studies of two methods of drug discover through to drug approval, despite what might be learned. The best that can usually be hoped for is to examine extant data and evaluate whether one method fits what was observed better than another. This requires an appreciation of statistics. However, the level of knowledge of what could be achieved is noticeably absent from presentations and publications. This can be addressed. The plan for this conference is to improve awareness of what is possible with modern statistics, to apply, so to speak, computation to computation. To do this, a four-pronged approach is planned. (1) To canvas for presentations of straightforward techniques from which any investigator might gain insight and which are less well-known than such value might imply. (2) To solicit work and presentations on major statistical issues in the field of CADD, namely the incorporation of error in both experimental data and computational process in the construction of models, methods to address over- parameterization of models, the assessment of the consequences of the non-ideality of retrospective datasets derived from the medicinal chemistry process, and the use and choice of appropriate NULL models in assessing new approaches. (3) To ask for allocutions from senior members of the field as to how statistical standards might be established and, finally, (4) presentations of perspectives from other disciplines. In addition, itis planned to have hands-on sessions where attendees may work with community data or bring their own and gain advice from experts. It is hoped that this conference will lead to regular self-assessment in the field of molecular modeling, standards for publication and presentations and help advance the rigorous discrimination of approaches to drug design and discovery.
Drug discovery and optimization is the most direct and translational application of physical science research to public health. Better understanding of when progress has been made in theoretical approaches to this endeavor is crucial for advancement. The conference proposed herein aims to improve this understanding.