Skip to main content
Fig. 6 | BMC Bioinformatics

Fig. 6

From: Increased peak detection accuracy in over-dispersed ChIP-seq data with supervised segmentation models

Fig. 6

The unconstrained model with a negative binomial or a Gaussian transformed noise is more accurate than previous state-of-the-art. The mean of differences in accuracy and its 95% CI computed on the test folds pooled by types of experiment (H3K36me3 & H3K4me3) are shown in red . If the mean of differences in accuracy is negative (left side of the blue indicator line ), the Up–Down model with a Poisson noise is better in average than the target model. The results of the paired t-test used to assess the difference of mean accuracy are summarized in the following way: non significant (ns) means adjusted p-value \(>0.05\); * means adjusted p-value \(\le 0.05\)

Back to article page