Fig. 3
Impact of L1-regularization on reduced feature sets. a-b Line-plots representing parameter tuning of regularizing term α against the mean-squared error (a, MSE) and the number of features remaining (b, p), across all cell cycle phases, where α∈[10−3,1]. Error bars indicate ±SD with 10-fold CV. c Bar plot representation of model coefficients (as importance) for each feature from the Gradient-boosted regression tree (GBRT), using optimized α, for each cell cycle phase. Error bars indicate ±SD with 10-fold CV. d Table of number of selected features per method, per cell cycle phase