Systems biology has faced challenges with dynamic mechanistic modelling due to the complexity and variability of the underlying relationships as well as the scarce and unclear experimental data.To address these problems, the statistical mechanics-derived idea of ensemble modelling has been applied to biological applications.
The phenomenon of interest is described using ensemble modelling, which combines a variety of models that are compatible with the data that have been observed.Ensembles of models, however, are particularly unreliable when forecasting non-observable states since systems biology models frequently struggle with a lack of identifiability and observability.
The authors outline a method for evaluating and enhancing a class of model ensembles’ dependability.They focus on kinetic models using ordinary differential equations that have a fixed structure.With their method, a subset of the parameter vectors discovered during parameter estimation using a global optimization metaheuristic are assembled into an ensemble.
This method can measure the uncertainty in state trajectory predictions and enforce diversity during parameter space sampling.When possible prediction problems are found via structural identifiability and observability analysis, they combine this method with model reparameterizations to address them
The end result is a group of models that can forecast how a biological process will behave internally.
The source code is available at the following URL: https://doi.org/10.5281/zenodo.6782638.
Reference:
Massonis G. et. al.(2023)Improving dynamic predictions with ensembles of observable models. Bioinformatics.39(1)btac755