Ns from the ENCODE cell lines H1-hESC (H1) and K562. We investigated H3K36me and H3K79me2 that are related to active transcription and occur preferentially in gene bodies. We also included the H3K9me3 data set in order to corroborate the results obtained on the mouse data, which had a relatively low read coverage (see Table 1). We were mainly interested to assess how versatile the compared methods are and to identify potential biases of any method towards certain histone modifications. For the evaluation we again compared the differentially called regions to differential gene expression, that was obtained from ENCODE RNA-seq data. Figure 3d shows that histoneHMM outperforms the other tools for H3K9me3 also in the ENCODE cell lines and thereby confirms the results based on the mouse data set. Figure 3e-f shows the performance for H3K36me3 and H3K79me2. It is worth noting that the relation between differential gene expression and differential histone modifications is much more pronounced for H3K36me and H3K79me2 than for H3K27me3 or H3K9me3 since the former are directly related to the transcriptional process. The results show that histoneHMM is an efficient algorithm for detecting functionally relevant differentially modified regions. This is likely due to an overall lower false positive and false negative rate. Indeed, extensive simulation studies support this conclusion (Additional file 1).Runtime evaluationWe evaluated the runtime of each algorithm on each of the data sets presented above. We measured the user time on a 1150 MHz Quad-Core AMD Opteron Processor 2356. Figure 4 shows that Chipdiff is the fastest algorithm on all data sets, followed by histoneHMM. Note that the figure has a log scale, so other algorithms are orders of magnitude slower.Heinig et al. BMC Bioinformatics 06:1)52(Page 6 ofaRat heart BN vs SHR H3K27mebMouse liver female vs male H3K9meDiffrepsDiffreps6.Chipdiff histoneHMM0.56 0.96 0.68 0.61 0.26 0 1.37 0 0.03 0 0.Chipdiff0 0 0.81 4.79histoneHMM0 0 0 0 0 0 5.61 8.36 105.09 2.85 2.011.0.07 0.2.0.52 0.0 0.12 0.06 2.23 1.002.24 0.0 0 0.2.22 0.13 0.29.1.2524.0.RsegPeprRsegPeprcENCODE H1-hESC vs K562 H3K9medENCODE H1-hESC vs K562 H3K36meDiffreps0.RsegChipdiff1321.0.histoneHMM14.09 0.3 51.3 19.3.64 0 0.03 0 0.0.28.0.09 0144.77 34.19.0.0 0 0 0 0.12 0.112.6 11.1.4112.36 33.23.1590.histoneHMMDiffrepsPeprRsegeENCODE H1-hESC vs K562 H3K79mefENCODE H1-hESC vs K562 H3K27meDiffreps Diffreps7.Chipdiff1.67 0.42.Chipdiff0.histoneHMM1.1.Rseg1390.histoneHMM56.61 5.64 74.5 22 0.59 15.52 0.15 22.99 0.03 0.0.218.14 0.16 31.71.0.17.0.08 0.01 0.03 0.16 17.0.1.51 1.0.086 2.12.97 1.95 42.30.15 308.39 18.1.44 34.42 12.3.1438.PeprRsegFigure 2 Venn diagram. The Venn diagrams show the overlap in Mb between the regions that were called differentially modified by each of the methods for the Entinostat chemical information analysis of a) strain differences of H3K27me3, b) sex differences of H3K9me3 and differences between ENCODE cell lines H1-hESC and K562 for c) H3K9me3, d) H3K36me3 (Chipdiff and Pepr did not identify any differential regions), e) H3K79me2 and f) H3K27me3 (Pepr did not identify any differential PubMed ID:https://www.ncbi.nlm.nih.gov/pubmed/26162776 regions).Application of histoneHMM to single ChIP-seq samplesAlthough histoneHMM was primarily designed for the detection of differentially modified regions between twoChIP-seq samples, it can also be easily applied to the analysis of a single ChIP-seq sample. In this case histoneHMM classifies the genome into regions that are modified orHeinig et al. BMC Bioinformatics.