Fig. 8
From: Genome-wide prediction of DNase I hypersensitivity using gene expression
BIRD predictions used as pseudo-replicates to improve DNase-seq and ChIP-seq data analyses. The observed signal from one sample (“obs-only”) in a test cell type was combined with BIRD predictions to produce the integrated signal (“BIRD + obs”). Signals before and after integration are compared with the observed signal from another sample from the same cell type (“truth”). a–c DNase-seq. a Correlation (r) between the “truth” and “BIRD + obs” (i.e., the integrated signal) in GM12878. Each dot is a genomic locus. b Correlation between the “truth” and “obs-only” (i.e., the original signal without integrating BIRD) in GM12878. c The same analyses were done for 16 test cell types. Red dots in the scatterplot compare the P–T correlation r for BIRD + obs vs. r for obs-only in the 16 cell types. BIRD + obs outperformed obs-only in 12 of 16 test cell types. As a control, BIRD was replaced by the mean DH profile of training cell types. Blue dots show the P–T correlation r for mean + obs vs. r for obs-only in the 16 test cell types. Mean + obs did not improve over obs-only. d–f ChIP-seq. d Correlation between the “truth” and “BIRD + obs” for SP1 in GM12878 at SP1 motif sites. e Correlation between the “truth” and “obs-only” for SP1 in GM12878 at SP1 motif sites. f The same analyses were done for nine TFs in GM12878 (circles) and three TFs in K562 (triangles). Once again, BIRD + obs outperformed obs-only in 9 of 12 cases (red and green), but mean + obs did not improve over obs-only (blue and yellow)
