Optimization of a signal transduction model of the mitochondrial apoptosis by training to multi-cell data.
Research in systems biology is a cycle of proposing and validating novel hypotheses through wet lab experiments and computational modelling. Hypotheses based on experimental data are tested in computational models and resulting model predictions propose further biochemical experiments which are once again basis for an improved and more broadly applicable model. Despite best experimental efforts and advances in biotechnology, essential parameters for deterministic models such as biochemical reaction kinetics are often not available in sufficient quality, and adapting these parameters through training of the model to control experiments remains the only alternative. Exemplified by our recently developed model of mitochondria1 apoptosis (I), this thesis presents a structural approach to broaden a computational model by training against single-cell data sets (2). Principal component analysis was used to identify sensitive and adaptable parameters in the model. Parameter optimization by classical Nelder-Mead fitting, a Monte-Carlo approach and a brute-force Screening were then iteratively applied to reduce ambiguity in candidate parameter values. Our approach leads to parameter sets enabling the model to correctly predict apoptotic kinetics of several cancer cell lines. 1. M. Rehm, H.J. Huber, H. Dussmann, and J.H.M. Prehn, The EMBO Journal, 2006, 4338-4349. 2. C.L. O'Connor, S. Anguissola, H.J. Huber, H. Dussmann, J.H. Prehn, and M. Rehm, Biochimica et Biophysica Acta (BBA) - Molecular Cell Research, 2008, 1903- 1913