Skip to main content
Fig. 4 | BMC Bioinformatics

Fig. 4

From: Extensions of 1 regularization increase detection specificity for cell-type specific parameters in dynamic models

Fig. 4

Parsimonious models for some choices of deformation parameters. Upper row: Parsimonious models by scanning over penalty strengths λ for a fixed deformation parameter. Only the log-fold-change parameters are shown. Filled circles depict the value of those log-fold-changes that are non-zero after regularization. Squares denote the 1 penalty at zero deformation, while circles represent the extended functions. Kinetic rates in bold face were chosen as cell-type specific for data simulation. The colours of circles after these kinetic rate names encode the corresponding value. For q, the parsimonious model was independent of the choice of q. Elastic Net did not induce an effective model shrinkage for α>0.5. Lower row: The total effective run times of the regularization routines as function of the deformation. These account for the entire two-step regularization routines to find the parsimonious models. All runs were executed on a 20-core server

Back to article page