Fig. 4: Description of application example, datasets, and model performance, when trained with mini-batch optimization.

a Simplifying illustration of the multi-pathway model of cancer signaling. b Left: Overview of the datasets used for training and model validation, taken from the Cancer Cell Line Encyclopedia. Right: Comparison of model sizes and experimental conditions used for model training of recently published ODE models. c Correlation of measured and simulated cell viability for all points of the training data, color-coding indicates density in scatter plot. d Receiver-operating characteristics for classification into responsive and nonresponsive combinations of cell lines and treatments on training data for the best ten optimization runs (gray) and an ensemble simulation (blue). e Area under ROC curve and classification accuracy on training data for the ten best optimization results (gray), for the ensemble model (black), and for the ensemble model on data for each drug individually (colored). f Simulated drug response. Left: Ranking of fit quality for cell lines by average root-mean-square error (RMSE). Right: Two out of 233 cell lines from the training data, error bars indicate the standard deviation across an ensemble of the n = 10 best optimization runs, for a cell line which the model was able to describe well (blue, BCPAP) and a cell line, which was less well captured by the model (orange, KU812).