Multiple linear regression was performed for all 25,144 expressed (max TPM >1) RefSeq TSSs with at least ten FANTOM5 CAGE-defined enhancers within 500 kb. Enhancers were ranked by proximity to the TSS and the expression values across all samples of the ten closest were used as predictor variables in a model with the TSS expression as response variable. The expression data of enhancers and TSSs were centered and rescaled. 2,206 TSS models, considering in total 11,386 enhancers, with R2 ≥ 0.5 were considered for further analyses. We also fitted a simple linear regression model using each enhancer as predictor variable on their own, in order to compare the predictive power of a single enhancer to the power of using all ten. We defined a new measure of ‘proportional contribution’ to the variance explained as the ratio between simple linear regression r2 and multiple linear regression R2, for each enhancer among the ten considered for each TSS. This measure yielded highly similar ranking results of enhancers as the R2 contribution averaged over orderings among regressors67,68 and R2 decorrelation decomposition67,69 (data not
