Graphical User Interface

OptiPa, an ODE modelling optimisation interface

​OptiPa tries to bring a user friendly interface to allow quick analysis of the developed models. This page shows some of the typical features of OptiPa.
(For a full feature list look here)

  • Stand/alone application As of version 6, OptiPa is provided as a compiled stand alone application. As a result, the user no longer has to have access to MatLab itself.

  • Model exchange To facilitate the exchange of models between users extra functionality has been created to export models into a packaged file format which can be easily exchanged and imported by another user. To protect you model code you can now encrypt a model, still allowing you to share the model with an end user without the end user being able to access the source code in a readable format.

  • Main screen After having prepared the relevant ASCI input files and the appropriate model file, the subsequent data analysis using the graphical user interface OptiPa  requires only some clicks of the mouse. To run an analysis one has to load the prepared model files, identify the experiments to be used, select the model parameters of interest, build a loss function and select an action to undertake. Each of the tasks can be activated from the button panel. On completion the user will be presented graphical, numerical and statistical outputs.

  • Observed-expected After a successful optimisation OptiPa returns graphical, numerical and statistical outputs. The graphical outputs consist of an overall plot of observed versus expected model values and a plot of residuals as a function of time. One can toggle between the two plot types by clicking the graphs.

  • Equation editor To facilitate the editing of the model equations OptiPa provides a simple equation editor that graphically represents  equations copied to the clipboard and, after editing, can be pasted back as an equation string into the model file or as graphics or LaTeX code to any other application.

  • Selecting your variables OptiPa tries to facilitate the whole modelling process through simple accessible interfaces. By ticking the boxes off the objective function window, measured experimental variables are linked to the model variables to tell OptiPa how to fit the model to the data. 

  • Simulations For each of the experiments OptiPa generates time course plots of the combined experimental and simulated data and of the residuals. Again, one can toggle between the two plot types by clicking the graphs. In these graphs data is plotted per experiment. By default, all dependent model variables are shown together. Per graph and per variable, data is scaled between its minimum and maximum value (the relative x and y-ranges going from 0-1). This allows quick evaluation of the model fit in case the different variables are of complete different magnitudes. By selecting a single dependent model variable from the listed variables at the bottom right corner, this selected variable will be shown using absolute scaling on both axis.

  • Goodness of fit One can toggle between the observed model values and the residuals by clicking on one of the graphs. When the residuals are plotted, the horizontal line represents the absolute zero line. In case of a good fit the residuals should be normally distributed around this zero line not showing any systematic deviations. The user is presented with statistics to judge the goodness of fit per experiment.

  • Optimisation methods OptiPa provides several ways of fitting the model to the data by estimating values for the unknown model parameters thus minimising the discrepancy between predictions by the model and the measured experimental data.  The least square optimisation is the default optimisation method but when the function has multiple local minima or one is very unsure about proper starting values for the various parameters one can use one of the genetic algorithms that all scan a much wider range of parameter values and therefore are potentially more successful in obtaining a global minimum.

  • Statistical output The statistical output has been designed in accordance with the standard output provided by the non linear regression routine from SAS (version 6.11, SAS institute Inc., Cary, NC, USA).
    After a listing of the iterative optimisation process, OptiPa calculates a summary table of the regression in terms of the sums of squares. It furthermore calculates the approximate standard errors, the 95 % confidence intervals and the correlation matrix of the estimated parameters based on the Jacobian matrix coming from the optimisation routine. Furthermore, using a 99 % confidence interval on the residuals the most extreme outliers are identified for each of the dependent variables.

  • Bootstrapping Given non linearity of the models, using standard statistics, confidence intervals for the model parameters can only be approximated. To acquire accurate estimates of the confidence intervals, bootstrap techniques are required and have been implemented in OptiPa. They allow the identification of asymmetric confidence intervals as is shown in the statistical analysis of the bootstrap results.

  • Conditional joined confidence intervals The parameter estimation accuracy of simultaneously estimated model parameters can be assessed by joined confidence regions. The joined confidence region is constructed by calculating the SSE for a grid of two-parameter combinations while keeping the other model parameters at their optimised (estimated) value (conditional joined confidence region). These confidence intervals allow a further identification of the asymmetric confidence intervals.

  • Monte Carlo Monte-Carlo simulation is a numerical stochastic technique used to solve mathematical problems. A Monte-Carlo simulation is based on some model system that can be described as a function of random model parameters characterised by their probability distribution functions. Monte-Carlo simulation simulates the model system after random sampling from these probability distribution functions. Monte-Carlo simulations can be performed to study model behaviour at different conditions taking into account the parameters' inaccuracies. This results in graphical and numerical output showing the confidence intervals for the simulated model output.