Charge optimization is an essential element of rational drug design; given ligand and receptor proteins, one wishes to determine the ligand charge distribution―a vector of partial atomic charges―that maximizes the favorable change in electrostatic free energy on binding. Work in biophysics has shown that this problem is convex, and that it can be solved using standard quadratic programming methods. However, the use of these techniques requires an initial calculation of the Hessian matrix; the present work introduces an new method that avoids this expensive computation and also scales much more favorably with problem size. The technique couples a boundary element formulation with a primal-dual interior point algorithm; initial results suggest that the cost to solve these optimization problems can be reduced by more than an order of magnitude.
展开▼
