Global Minimization and Parameter Estimation in Computational Biology

Two important problems in computational biology are the prediction of the structure of a protein, given its sequence of amino acids (the protein folding problem), and the docking of a ligand on a protein (drug design). Both of these problems require the computation of the global minimum (or near global minimum) of a differentiable function with many local minima. The Convex Global Underestimator (CGU) algorithm for solving this type of problem has been developed and implemented and successfully tested on a variety of structure prediction and docking problems. The function representing the potential energy of a protein molecule in terms of its dihedral (backbone) angles (a vector Φ)) has approximately 2n independent variables for a protein consisting of n amino acids.