Supporting information – Text S5: Wild-type model analysis A computational model for the analysis of lipoprotein distributions in the mouse: Translating FPLC profiles to lipoprotein metabolism

A computational model for the analysis of lipoprotein distributions in the mouse: Translating FPLC profiles to lipoprotein metabolism Following parameter estimation and evaluation, two parameter sets were found to describe the data well (these will be referred to as sets X1 and X2). Inspection of the in silico FPLC profiles (e.g. Figure 1 in Text S4) reveals that in the VLDL and LDL size range, the profiles of X1 and X2 do not overlap. For the HDL size-range, however, the profiles are very similar. Keeping this observation in mind, the parameter sets themselves (Table 2 in Text S4) only partly collaborate these observations. The VLDL sub-model parameters indeed show great differences, e.g. in the production diameter and particle number (a factor of two difference in # of produced particles) as well as catabolism. The lipoprotein shape-dependent parameters (which are linearly scaled), correspond fairly well between parameter sets. This indicates the general structure of VLDL remodelling is similar between the two parameter sets. For further analysis of the two parameter sets, we performed re-sampling of the found optimal parameter sets and calculated a profile likelihood (similar to [9], starting from both initial sets), as well as error plots of VLDL sub-model parameters. Parameter space and parameter identifiability Re-sampling of the optimal parameter sets For a general idea of the parameter space near the optimal parameter sets, we re-sampled around the optimal sets. This was done be creating parameter sets with between-10 % and 10% and between-20% and 20% uniformly distributed random noise around the optimal parameter sets. Each set, for both 10% and 20%, was re-sampled 140 > times. A total of 731 re-samplings were performed. All re-sampled parameter sets were optimized as described in Text S4. Analysis of the results revealed that again two types of acceptable results (according to the criteria defined in Text S4) were found, which could be classified as "X1" or "X2". Furthermore initial sets derived from "X1" were sometimes re-optimized to "X2" and vice versa. A basal profile likelihood was performed. The object of the profile likelihood for parameter x is to find optimized values of all other parameters for a range of values of x and investigate for which x an acceptable result is attained, i.e. to investigate the parameter identifiability of the parameters in the model (e.g. [9]). The profile of parameter x is initialized with the optimal set opt  (either X1 …