Parameter optimization in differential geometry based solvation models

摘要

Differential geometry (DG) based solvation models are a new class ofvariational implicit solvent approaches that are able to avoid unphysicalsolvent-solute boundary definitions and associated geometric singularities, anddynamically couple polar and nonpolar interactions in a self-consistentframework. Our earlier study indicates that DG based nonpolar solvation modeloutperforms other methods in nonpolar solvation energy predictions. However,the DG based full solvation model has not shown its superiority in solvationanalysis, due to its difficulty in parametrization, which must ensure thestability of the solution of strongly coupled nonlinear Laplace-Beltrami andPoisson-Boltzmann equations. In this work, we introduce new parameter learningalgorithms based on perturbation and convex optimization theories to stabilizethe numerical solution and thus achieve an optimal parametrization of the DGbased solvation models. An interesting feature of the present DG basedsolvation model is that it provides accurate solvation free energy predictionsfor both polar and nonploar molecules in a unified formulation. Extensivenumerical experiment demonstrates that the present DG based solvation modeldelivers some of the most accurate predictions of the solvation free energiesfor a large number of molecules.