Skip to main content

Advertisement

Fig. 4 | BMC Systems Biology

Fig. 4

From: Inference of spatiotemporal effects on cellular state transitions from time-lapse microscopy

Fig. 4

Regularized generalized linear models (GLMs) select the relevant features predictive of cell state transitions. a Regularization path of the GLMs applied to the density dependent dataset. The means (lines) and standard deviations (shaded regions, shown only for the relevant features) of the regression weights w are plotted against the regularization strength κ across 50 bootstrap samples (see Methods for details). The mean of the optimal regularization strength κ determined by cross validation is shown as a vertical black line. Solid (dashed) lines correspond to relevant (irrelevant) features in the respective scenario. b Percentage of bootstrap samples that included the respective features. Included features were determined as those with non zero weights at κ . Enforcing a 90 % threshold (gray area) on the inclusion probability for each feature, we select the relevant features of the model. The features ϕ 0,ϕ 1 are not included as their effect is too weak to be detected by the GLM at the current sample size (see main text). c Reconstructed kernel of local cell density (bars) from the selected features in b. The true underlying tophat kernel shape is shown in black. As in b, the features ϕ 0,ϕ 1 are not included because their effect is to weak. d-f Analogous to a-c, but for a dataset where the transition rate λ depends on time and local cell density with a Gaussian kernel. Both features are correctly identified and the density kernel is correctly estimated

Back to article page