We extended the relevant biological assays with two computational predictors. First, we predict TF binding in a cell-type–specific manner by predicting DNase footprints from the DNase-seq data sets (Boyle et al. 2011; Pique-Regi et al. 2011). These results indicate a specific DNA–protein interaction event and, when combined with variant information, increase the confidence that a SNV in this region is functional. Secondly, we scanned the genome at a reasonable threshold for added positional weight matrices (PWMs) (Berger et al. 2006, 2008; Matys et al. 2006; Bryne et al. 2008; Badis et al. 2009; Scharer et al. 2009; Wei et al. 2010), which were further used to intersect with other functional data (see Methods). For these computational predictions of protein binding, 1158 motifs were considered, which resulted in over 365 million additional annotations in RegulomeDB.
