Figure 3

Method overview: four stages in the investigation of the high-dimensional non-linear dynamic model: (1) Preliminary exploration. Design: trial-and-error. Analysis: visual inspection of model steady-state solutions, rendered as black and white images of Notch concentrations. (2) Large-scale systematic screening using extensive probing of the parameter space, monitored by computerised profiling by full factorial designs. Relating more than 50 computed image analysis descriptors to 7 known model parameters by regression in latent variables (jack-knifed PLSR). (3) Disciplined investigation using informative but more laborious human sensory descriptive profiling by a reduced factorial design, and relating 15 sensory image descriptors to all 7 model parameters by jack-knifed PLSR. (4) Pursuit of an unexpected discovery in a particular region of parameter space. Dense sampling of θ_D for chosen combination of other parameters, and relating estimated protein distributions and sensory image descriptors to θ_D by graphics.