DescriptionComputational modelling in systems biology addresses biological processes at different levels and scales. The quantification of model parameters from experimental data is a complicated task. To develop accurate, predictive models it is necessary to analyze how variance in data propagates into parameter estimates and, more importantly, model predictions. The network structure of the biological systems imposes strong constraints on possible solutions of a model. Amounts of data, available at molecular and physiological level, continue to increase. Often, model results are only partly in agreement with data, despite that model parameters are fitted. In contrast to existing belief that calibration of systems biology models to experimental data is prone to overfitting, we argue that dynamical models, despite their size and complexity, are not flexible enough to correctly describe all data. Approaches are explored to introduce more degrees of freedom in models, but simultaneously enforcing sparsity if extra flexibility is not required. Estimation tools for dynamical systems are complemented with ‘regularization’ methods to reduce the error (bias) in models without escalating uncertainties (variance). This paradigm shift will be illustrated in two examples: 1) modelling of longitudinal data in a cohort of Type 2 Diabetics using different medication, and 2) the application in preclinical research studying the effect of liver X receptor activation on HDL metabolism and liver steatosis.
|Period||20 Apr 2016|
|Event title||2nd Dutch Bioinformatics & Systems Biology Conference (BioSB 2016), April 19-20, 2016, Lunteren, The Netherlands|
- systems medicine
- parameter estimation
- uncertainty analysis