Global optimization plays an important role in computational molecular biology. In computational protein folding and docking with direct applications to drug design, for example, a principal problem is that current conformational search methods are not efficient enough at finding the deep low-energy states on the energy landscapes. The recently developed general Convex Global Underestimator (CGU) method provides dynamic and flexible under-fitting of the energy landscape. It is not hindered by kinetic traps in its search speed, and has been proved successful on docking. Based on all-atom model energy landscapes, we propose new conformational search strategies that combine the general CGU with a physics-based """"""""zipper method"""""""" for folding. Our search methods can be efficiently applied to a wide variety of energy profiles. We will test the new search strategies on some small proteins to see if and how often the native states can be reached in all-atom models.
