Differences in expression prediction accuracy between dissimilar CRMs. Different models, one generated from GM12878 CRMs and the other from K562 CRMs, were used to predict gene expression levels in the same cell type (K562). The ratio between the mean squared prediction errors of the GM12878 model (MSPE.GM12878) and K562 model (MSPE.K562) is higher for CRMs where a larger proportion of TF binding sites (Jaccard distance) differ between the two cell types.
Wang et al. BMC Genomics 2012 13:263 doi:10.1186/1471-2164-13-263