A program for the identification of steady-state processes in biological systems

A dedicated nonlinear regression program for the identification of steady-state processes is described in detail. Experimental data are fitted to the rational function which describes such systems. The algorithm makes use of parameter separation to relieve the user from the need to assign initial estimates to the linear parameters and to speed up the computation. A modified Marquardt algorithm is used and some properties of the function are exploited to improve the convergence rate and execution time. The search is made constrained by the addition of a penalty function to the residual sum of squares. Graphics are included to help the user find good initial estimates for the nonlinear parameters and as an aid to judge the goodness of fit of the result. Residual plots are also available to help the user identify the origin of the differences in the statistical tests. A final prediction error test is provided since this statistic is thought to be the most appropriate for the nonlinear function used. An enzyme kinetic example illustrates the procedure. The program was coded in Turbo Pascal on an IBM compatible PC.