Validation of systems biology models

The paradigm shift from qualitative to quantitative analysis of biological systems brought a substantial number of modeling approaches to the stage of molecular biology research. These include but certainly are not limited to nonlinear kinetic models, static network models and models obtained by the analysis of large scale datasets such as clusters in gene expression data or principal component analysis models. However, the concept of 'model validation' is not encountered as often as the introduction of new models in the field. This leaves many of the proposed models untested and therefore, creates a gap between the number and the reliability of the models.
With this thesis, we present computational approaches for model validation and provide guidelines for reliable model validation and selection with examples on real biological data.
The second and third chapters of this thesis focus on nonlinear kinetic models by which we can model the dynamics that underlie processes within a cell. They are usually formulated by using sets of ordinary differential equations (ODE) with many unknown parameters. Most of the time, there are competing hypotheses on the biochemical species, the regulatory relations that these models contain and the governing biochemical rules. This uncertainty brings the need for careful model selection and model validation.
The fourth and the fifth chapters focus on the validation of two conceptually different types of models, both regarding transcriptional regulation. In the first type, a regulatory transcriptional network model is used to summarize the physical association between genes and transcription factors. In the second type, clusters obtained from a cluster analysis summarize the similarity of the expression profiles of genes across various arrays.