Skip to main content
Fig. 2 | Cell Communication and Signaling

Fig. 2

From: L-plastin Ser5 phosphorylation is modulated by the PI3K/SGK pathway and promotes breast cancer cell invasiveness

Fig. 2

Computational modelling approach. a Inference of the cell line-specific parameters: (1) input data consist of a generic network topology and measurements of the activation state of output nodes (ratio phosphoprotein/total protein) for four cell lines, each cell line being represented by a different colour. (2) Without any regularization, model parameters are independent across cell lines, which might result in overfitting of the dataset. (3) When, in contrast, model parameters are forced to be equal across all cell lines, the phenotypes of the different cell lines are smoothed out and only the average behavior can be inferred. (4) By applying various levels of regularization (penalizing model size), the sparsity of the model, i.e. the number of model parameters allowed to vary across cell lines, can be controlled. (5) The Bayesian Information Criterion (BIC) is a measure of adequacy of the model, balancing model fitness and model size. The model configuration with the lowest BIC, evidencing the most crucial differences in signalling between the cell lines, was chosen for the final analyses. b Regularization landscape. Heatmap of the BIC values for each regularized model. Optimal model parametrization was obtained by screening values of the lambda_Pruning and lambda_Uniformity regularization parameters and computing the BIC for each resulting model. The model with the lowest BIC is considered the most adequate to represent the data and is obtained with log2(lambda_Pruning) = − 10 and log2(lambda_Uniformity) = − 4. c Best fit. Comparison of the simulated activity of the different measured proteins (output nodes) with the measurements. X-axis: different experimental conditions. Y-axis: normalized activity. Green: average of experimental measurements. The error bars represent 1 standard deviation. Blue: activity as simulated with the FALCON toolbox under the optimized final model topology

Back to article page