Modelling of the breadth of expression from promoter architectures identifies pro-housekeeping transcription factors
Fig 7
The agreement between the models is expressed in the correlation between the vectors of coefficients produced by different models. This figure shows the matrix of scatterplots for pairwise comparisons between: the binomial GLM, lasso, ridge, the elastic net, the Bayesian model, together with the three cluster (I, II and III) of the non-linear-response model. The panel above the diagonal shows pairwise correlation coefficients (Spearman correlation) between the vectors of regression coefficients βi. On the diagonal, a histogram of coefficients from each of the models is displayed. The classical GLMs are very similar to each other (with Spearman’s rho between 1 and 0.97). The Bayesian model is slightly more diverged (rhos between 0.92 and 0.93).