Differential peak calling of ChIP-seq signals with replicates with THOR

Abstract

The study of changes in protein-DNA interactions measured by ChIP-seq on dynamic systems, such as cell differentiation, response to treatments or the comparison of healthy and diseased individuals, is still an open challenge. There are few computational methods comparing changes in ChIP-seq signals with replicates. Moreover, none of these previous approaches addresses ChIP-seq specific experimental artefacts arising from studies with biological replicates. We propose THOR, a Hidden Markov Model based approach, to detect differential peaks between pairs of biological conditions with replicates. THOR provides all pre- and post-processing steps required in ChIP-seq analyses. Moreover, we propose a novel normalization approach based on housekeeping genes to deal with cases where replicates have distinct signal-to-noise ratios. To evaluate differential peak calling methods, we delineate a methodology using both biological and simulated data. This includes an evaluation procedure that associates differential peaks with changes in gene expression as well as histone modifications close to these peaks. We evaluate THOR and seven competing methods on data sets with distinct characteristics from in vitro studies with technical replicates to clinical studies of cancer patients. Our evaluation analysis comprises of 13 comparisons between pairs of biological conditions. We show that THOR performs best in all scenarios.

Figure 1.

THOR's analysis workflow. (A) After pre-processing the ChIP-seq signal, normalization based on housekeeping genes (as shown) or TMM is performed. The normalized signal serves as input for the HMM which is used to estimate DPs. Post-processing includes the statistical assessment of DPs. (B) List of all competing methods categorized in one-stage and two-stage approaches. (C) Evaluation strategies. We evaluate DPC methods with biological data and the DCA statistic that is based on the association between DPs, gene expression and histone modifications. Moreover, simulated data are used to investigate the effect of distinct ChIP-seq signal characteristics on DPCs methods.

Figure 2.

Results for simulated data. We show the AUC distribution for 25 repetitions of each scenario. Simulated data were based on (A) moderate and (B) high condition peak size variability and 2 (red lines) and 4 (green lines) replicates. Each boxplot is divided by the level of within condition variance (low, medium and high). Methods (x-axis) are ordered by decreasing median AUC values for the cases with 4 replicates.

Figure 3.

Association between average FRIP and overdispersion scores α. (A–B) We show the relation between mean and variance of replicates on two selected experimental conditions (LYMP-CC) and (DC-cDC). (C) FRIP and overdispersion scores for the 26 biological conditions analysed: cocaine intake (CO), monocyte differentiation (MM), lymphoid cancer (LYMP) and dendritic cell differentiation (DC). Higher FRIP indicates higher signal-to-noise ratio and better ChIP-seq experiments. Higher overdispersion scores indicates higher within condition variability.

Figure 4.

DCA curves for selected DP problems. We show DCA curves for peaks detected in H3K4me1 of the Cocaine Response study (CO-H3K4me1), which were evaluated by (A) RNA-seq and (B) H3K36me3; or H3K27ac peaks of the Lymphoma study (LYMP-FL-CC-H3K27ac), which were evaluated by (C) microarrays and (D) H3ac histone. Higher DCA values indicate higher association between differential peaks and differential expression or validating histones.

Figure 5.

Example of differential peaks. We depict H3K4me3 and RNA-seq signals for monocytes (red) and macrophages (green) around the promoter of (A) IRAK3 and PDK2; (B) H3K27ac signals around CRCX4 for Follicular Lymphoma (FL) and control (CC) individuals; and (C) Pol2 signal around Dvl1 for mouse treated with cocaine (green) and saline (red); Below the ChIP-seq signals, we depict differential peaks of all evaluated methods. Methods that do not detect peaks for a given experiment are not listed.