• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of bioinfoLink to Publisher's site
Bioinformatics. Jun 1, 2011; 27(11): 1569–1570.
Published online Apr 5, 2011. doi:  10.1093/bioinformatics/btr165
PMCID: PMC3102220

DIME: R-package for identifying differential ChIP-seq based on an ensemble of mixture models

Abstract

Summary: Differential Identification using Mixtures Ensemble (DIME) is a package for identification of biologically significant differential binding sites between two conditions using ChIP-seq data. It considers a collection of finite mixture models combined with a false discovery rate (FDR) criterion to find statistically significant regions. This leads to a more reliable assessment of differential binding sites based on a statistical approach. In addition to ChIP-seq, DIME is also applicable to data from other high-throughput platforms.

Availability and implementation: DIME is implemented as an R-package, which is available at http://www.stat.osu.edu/~statgen/SOFTWARE/DIME. It may also be downloaded from http://cran.r-project.org/web/packages/DIME/.

Contact: ude.uso.tats@ilihs

1 INTRODUCTION

ChIP-seq, chromatin immunoprecipitation (ChIP) assay followed by sequencing technology, is gaining popularity as the experiment to analyze genome-wide protein–DNA interactions at high resolution. ChIP-seq experiment produces millions of short sequences that needs to be aligned to a reference genome. The location where a significant number of short reads overlapped on the genome (sometimes referred to as a peak) has been shown to coincide with chromatin immunoprecipitation enrichment, indicating the binding site of the protein of interest.

A number of methods have been proposed to identify peaks in ChIP-seq data, such as FindPeaks, MACS and CisGenome. For more details on these and other existing algorithm including their comparisons, see Laajala et al. (2009). These softwares focus on identifying the peaks in one sample or in comparison with a matching input DNA. Whereas in DIME, we focus on identifying differential binding sites of a specific protein under two different biological conditions. The input to DIME is normalized differences of ChIP-seq counts, as in our recent work (Taslim et al., 2009). Specifically, in that paper, we proposed a method to normalize and classify the enrichment regions that are significantly different between two ChIP-seq samples (Taslim et al., 2009). Since then we have thoroughly compared the fitting of ours with another mixture model in the literature (Dean and Raftery, 2005), proposed an ensemble approach to synthesize advantages from different approaches and developed an R-package (DIME) to implement this method. By ensemble, we mean using a collection of three classes of mixture models where the best overall model is selected using BIC and AIC criteria (Hastie et al., 2009). Based on the best overall model, we classify each observation using local FDR (Khalili et al., 2009). One class of models used is the Normal-Uniform model (NUDGE) (Dean and Raftery, 2005). We have modified this class to improve the fit by allowing for multiple normals, leading to the class of iNUDGE models. The third class of models considered is the Gamma-Normal-Gamma (GNG) mixture (Khalili et al., 2009), which was adopted in our recent ChIP-seq analysis (Taslim et al., 2009). In GNG, the differential sites are captured by two exponential components as opposed to a uniform distribution as in iNUDGE. Further, for GNG and iNUDGE, some of the normal components may represent differential sites as well. As such, the applicability of DIME is greatly extended beyond uniform or exponential, since any distribution can be well approximated by a mixture of normals. By utilizing multiple models in a single analysis, we are able to improve the fit of the model to the data, which in turn improves the performance of the differential analysis for data from various omic platforms.

2 DESCRIPTION OF R FUNCTIONS

The main function of the package is DIME, which performs model fitting followed by classification of differential binding sites using normalized differences of ChIP-seq counts. Below we briefly describe its usage and functions.

Input: the only required input is an R list that contains normalized differences for chromosome(s) that need to be analyzed. Each element of the list contains data from one chromosome. Users can conveniently include/exclude chromosome(s) to/from the list depending on which chromosome(s) are of interest. The following would be an example of a call to the main function DIME with some optional parameters:

An external file that holds a picture, illustration, etc.
Object name is btr165i1.jpg

where data is an R list as described above.

Optional input parameters are available to control the fit and classification process. In the example above, two convergence criteria can be specified by users: gng.max.iter assigns the maximum number of iterations for fitting GNG model (default = 2000) and gng.tol (default = 1e−5) specifies the L2 norm of differences in the GNG parameter estimates in the current and previous iterations. Thus, the algorithm will stop whenever either one of these two criteria is satisfied. The maximum number of normal components searched when fitting GNG model is set using gng.K (default = 2). User can also adjust the local FDR for GNG-based classification by setting gng.fdr.cutoff (default = 0.1). Similar optional input parameters are available for the other two classes of models. Other additional optional input parameters are discussed in details in the manual of the R package.

Output: the output of DIME is a list containing four elements each corresponding to the results of the overall best model or those of an individual class. The results include details about the fit and classification, such as the estimated parameters for each component of the mixture and mixing proportions. For example, if the best overall model is GNG, then result$best = result$gng. Thus, result$best$pi gives the estimated proportion of each components under the GNG model. Further, result$best$mu and result$best$sigma provide the estimated means and SDs, respectively, for the normal components under GNG. A complete list of all available parameters and their descriptions are described in the user guide.

Our package also provides a number of graphical functions that produce plots for the best model as well as for the three individual classes. In the next section, we will give examples of these functions which were used to produce the figures displayed here.

3 EXAMPLE

To demonstrate the utility of the package, DIME was run to compare ChIP-seq data of a normal breast cancer cell line (MCF7) before and after Estradiol (E2) treatment with Polymerase II antibody (Taslim et al., 2009). The models were estimated using the default parameters, except for the maximum number of normal components which was set to be 5. Thus, the command used to run the analysis was as follows: result < −DIME(data,gng.K = 5, inudge.K = 5), where data are the normalized difference of ChIP-seq counts before and after E2 treatment (dataset included in the R package). It took 4675 s on an AMD quad-core 2.4 GHz processor to fit all three classes of models (searching up to five normal components) on around 20k genes (data points) using one random seed. As with any other random search algorithm, the running time of DIME is dependent on how good the initial parameters are. Figure 1A depicts the QQ-plot of the observed data against iNUDGE, GNG and NUDGE (from left to right), which are generated using inudge.plot.qq, gng.plot.qq, and nudge.plot.qq, respectively. The program selected GNG as the best overall model as it provides the best fit compared with iNUDGE and NUDGE, which is evident in Figure 1(A). Figure 1(B) shows the GNG density plot along with its individual components superimposed on the histogram of the normalized data, generated using the gng.plot.fit function.

Fig. 1.
(A) From left to right shows QQ-plot of iNUDGE, GNG and NUDGE. (B) Plot of the best mixture model (GNG) superimposed on the histogram of normalized data. N(.,.) stands for normal component with mean and SD. E(.) stands for exponential component with its ...

4 DISCUSSION

We have developed an R package (DIME) to model and make inference on ChIP-seq experiments under two different conditions. The algorithm effectively selects the model that provides the best fit to the normalized data, which lead to statistical inferences with high sensitivity and specificity. DIME can be easily combined with other R or Bioconductor packages to perform upstream and downstream analysis of ChIP-seq data. Furthermore, even though DIME is developed to fit and classify ChIP-seq data, it is highly applicable to other high-throughput data as well especially given its ensemble nature.

Funding: This work was supported in part by National Cancer Institute (grant U54CA113001) and National Science Foundation (grant DMS-1042946).

Conflict of Interest: none declared.

REFERENCES

  • Dean N., Raftery A.E. Normal uniform mixture differential gene expression detection for cDNA microarrays. BMC Bioinformatics. 2005;6:173. [PMC free article] [PubMed]
  • Hastie T., et al. The elements of Statistical Learning: Data Mining, Inference and Prediction. 2. New York: Springer; 2009.
  • Khalili A., et al. A robust unified approach to analyzing methylation and gene expression data. Comput. Stat. Data Anal. 2009;53:1701–1710. [PMC free article] [PubMed]
  • Laajala T., et al. A practical comparison of methods for detecting transcription factor binding sites in chip-seq experiments. BMC Genomics. 2009;10:618. [PMC free article] [PubMed]
  • Taslim C., et al. Comparative study on chip-seq data: normalization and binding pattern characterization. Bioinformatics. 2009;25:2334–2340. [PMC free article] [PubMed]

Articles from Bioinformatics are provided here courtesy of Oxford University Press
PubReader format: click here to try

Formats:

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

  • PubMed
    PubMed
    PubMed citations for these articles

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...