Methodological performance reporting: To confuse or to enlighten

COMP 147

Ajay N. Jain, ajain@jainlab.org, Cancer Research Institute, University of California, San Francisco, Box 0128, San Francisco, CA 94143-0128
Methods in molecular modeling, including docking, ligand similarity computation, and 3D QSAR all aim to address substantive operational issues in drug discovery. If the purpose of publishing papers and giving presentations on methods is to communicate the effectiveness of methods on the likely operational performance of the methods, there are implications for test set composition and preparation, computational procedure specification, and statistical reporting. Different approaches will be presented in the context of whether specific choices lead to confusion or to enlightenment. Agreement to follow approaches that lead to enlightened understanding of real-world performance will improve the field by accelerating scientific discovery, and will lead to more effective tools whose operational performance matches the claims of methods developers.