Uncertainty and Ensemble Modelling

Uncertainty modelling

Acknowledging that uncertainties can surround kinetic parameter values such as the difference between the values that are experimentally measured with the ones reported in literature, we take into account all known kinetic parameter values that can be found for each enzymatic reaction in our engineered metabolic network. These known values are then processed through our parameter algorithm that weighs each data according to its measurement condition and its suitability to our modelling system. The heavier the value within this set of data is weighted, the more plausible it is as the kinetic value for its respective kinetic parameter. This weighted values can then be analysed to determine its mode when generating probability distributions. Probability distributions can be seen as priors or our belief of what are the values that the kinetic parameter can take. For cases where only a single value can be found for a kinetic parameter, the value is multiplied and divided with a factor of 10 to cast a wide enough confidence interval that would include all plausible values in its probability distribution. Each enzymatic reaction which is modelled using the reversible Michaelis-Menten equation may require two or more kinetic parameters. Hence, each reaction in the model requires two or more probability distributions to represent each of its kinetic parameters. Sampling from each of these probability distribution would provide a set of kinetic parameter values and this means that the model can have multiple sets of kinetic parameter values for each reaction. Obviously, modelling the reaction with different sets of kinetic parameter values would provide different set of predictions and this is where ensemble modelling is used in this study.

Ensemble modelling

Instead of running a single prediction with a single set of kinetic parameter values, the metabolic model is run a number of times from different set of kinetic parameter values, which are sampled from probability distributions. The complete set predictions is referred to as the ensemble and individual predictions within it as predicted samples.

You can go back to main page of the kinetic model here.