Journal: Bioinformatics

Article Title: Optimization and profile calculation of ODE models using second order adjoint sensitivity analysis

doi: 10.1093/bioinformatics/bty230

Figure Lengend Snippet: Performance measures of different local optimization methods [ lsqnonlin with TR algorithm and fmincon with TR and IP algorithm, using either Hessians (H), FIM, or the BFGS scheme]. The multi-start optimization was carried out multiple times using different starting points for the local optimizations. Mean and standard deviation for ( A ) the ratio of computation time over converged optimization starts and ( B ) the number of converged starts are shown. ( C ) Median and standard deviation of the number of optimization steps over all optimization runs

Article Snippet: Common least-squares algorithms such as the MATLAB function lsqnonlin only use first order derivatives of the residuals, whereas constraint optimization algorithms like the MATLAB function fmincon exploit first and second order derivatives of the objective function.

Techniques: Standard Deviation

Journal: Bioinformatics

Article Title: Optimization and profile calculation of ODE models using second order adjoint sensitivity analysis

doi: 10.1093/bioinformatics/bty230

Figure Lengend Snippet: Profile likelihood computation using either the optimization-based method ( lsqnonlin or fmincon with TR or IP algorithm and Hessian or BFGS approximation), or the hybrid method with either FIM or Hessian. ( A ) Total computation time for all profiles of the considered models. Three methods failed to compute profiles for the RAF/MEK/ERK model. Thus, their computation times are not depicted. ( B ) A profile of the RAF/MEK/ERK model, the three remaining methods in good agreement with each other, with confidence intervals to a confidence level of 95% depicted below. ( C ) A profile of the RAF/MEK/ERK model, for which lsqnonlin failed to compute the profile, with confidence intervals to a confidence level of 95% depicted below

Article Snippet: Common least-squares algorithms such as the MATLAB function lsqnonlin only use first order derivatives of the residuals, whereas constraint optimization algorithms like the MATLAB function fmincon exploit first and second order derivatives of the objective function.

Techniques: