Accurate and Efficient Solution of Linear and Nonlinear ME Models

Integrated models of metabolism and macromolecular expression are multiscale and increasingly large. For linear ME models, double-precision simplex or barrier solvers can get reasonably close to the solution but they are not reliable. To obtain sufficient accuracy at reasonable cost, we warm start a quadruple-precision simplex solver (quadMINOS). For nonlinear ME models where the growth-rate is a variable, the same approach can be used with binary search (solving a sequence of linear models with growth-rate fixed at increasingly better values). Instead, we apply Double then Quad versions of MINOS to solve the problem as a single nonlinear model.