We propose to improve the speed and efficiency of the Convex Global
Underestimator (CGU) method, a powerful and relatively new global optimization tool, for its application to conformational search in energy-landscape approaches to protein folding. The method avoids becoming trapped into parts of the landscape that correspond to relatively low-energy, yet misfolded structures, a common problem for other continuous-conformation-space search methods such as molecular dynamics, Monte Carlo search, and simulated annealing. We will introduce three specific modifications to the CGU method in order to enhance performance. In the first approach, we will interface a ìfront-endî algorithm with CGU that can generate initial sets of low-energy conformations which allow the CGU method to improve its initial underestimator surface. In addition, we will explore means of reducing the degrees of freedom in our protein model without compromising chain flexibility. This reduction will correspondingly reduce the dimensionality of the energy landscape, decreasing the computational burden on the CGU method. Finally, we will use existing conformational data and established means of exploring energy landscape topographies to construct new underestimator surfaces that better approximate the protein landscape funnel and could augment the accuracy and convergence speed of the CGU method. We introduce these improvements with the intention of moving the method from its initial success with small test proteins to applications involving potentially much larger proteins and otherbiomolecules. |
