Contents

1 Preparations

We first set global chunk options and load the packages that are necessary to run the vignette.

library(DChIPRep)
library(ggplot2)
library(DESeq2)

2 Introduction to the DChIPRep package

The package implements the analysis strategy of (Chabbert et al. 2015). Along with an experimental protocol to multiplex ChIP–Seq experiments, (Chabbert et al. 2015) developped a methodology to assess differences between chromatin modification profiles in replicated ChIP–Seq studies that builds on the packages DESeq2 and fdrtool. The package also includes a preprocessing script in Python that allows the user to import sam files into data structures than can be used with the package.

3 Introduction to the pre–processing script

Together with the release of the R package DChIPRep, we provide a python framework that should allow users to generate count tables. These may then be directly used to evaluate metagene enrichment profiles. This framework is entirely contained in a script that only requires the installation of Python 2.7 and of the packages HTSeq and Numpy, which are free to download. This section describes briefly the various possibilities of analysis offered by this tool together with the various input options and parameters. Additional information may be found directly in the source code or via the usual help framework offered by Python.

The script comes with the package and can be located on your system by the following commands:

pythonScriptsDir <- system.file( "exec" , package = "DChIPRep" ) 
pythonScriptsDir
## [1] "D:/biocbld/bbs-3.2-bioc/tmpdir/Rtmpi60aeR/Rinst9c45d7c5833/DChIPRep/exec"
list.files(pythonScriptsDir)
## [1] "DChIPRep.py"

3.1 General principle

This script will process alignments of paired-end reads, filter out divergent pairs and low quality alignments and ultimately identify potential PCR duplicates. For the pairs of read retained after filtering, the center of the genomic loci determined by each pair is estimated using mapping coordinates. This center constitutes an approximation of the center of each observed nucleosome. The provided annotation file is then used to estimate the nucleosome counts around the features of interest. These counts are ultimately used to provide a count table in which each row corresponds to one feature and each column to a genomic position around the feature start. This table may then be imported in R using the importData function of the DChIPRep package.

3.2 Quick start

Here we show how to quickly generate a count table from an alignment file. The DChIPRep.py script only requires 4 arguments (all other options have a default setting, cf below):

Here is an example of the command line to be run

python DChIPRep.py -i alignment.sam -a my_gff.gff -g chromosome_sizes.txt -o my_count_table.txt

3.3 Advanced options

In order to provide greater flexibility in data pre-processing, we have added a panel of additional options that may be specified when running the script. The script is executable, With the –help option, one can get an overview of the available options. They are given below.

usage: DChIPRep.py [-h] -i SAM/BAM -a GFF -g Chromosome Sizes File -o Count Table [-v] [-q TH_QC] [-l TH_LOW] [-L TH_HIGH] [-d TH_PCR] [-f FEATURETYPE]
                   [-w DOWNSTREAM] [-u UPSTREAM]

optional arguments:
  -h, --help            show this help message and exit
  -i SAM/BAM, --input SAM/BAM
                        The alignment file to be used for to generate the count table. The file may be in the sam (zipped or not)format. The extension of
                        the file should contain either the '.sam' indication. Bam files are not supported at the moment due to soome instability in the BAM
                        reader regarding certain aligner formats.
  -a GFF, --annotation GFF
                        The annotation file that will be used to generate the counts. The file should be in the gff format (see
                        https://www.sanger.ac.uk/resources/software/gff/spec.html for details).
  -g Chromosome Sizes File, --genome_details Chromosome Sizes File
                        A tabulated file containing the names and sizes of each chromosome. !!! The the chromosome names should be identical to the ones
                        used to generate the alignment file !!! The file should look like this example (no header): chromI 1304563 chromII 6536634 ...
  -o Count Table, --output_file Count Table
                        The output file where the count table should be stored. If the specified file does not already exist it will be created
                        automatically. Otherwise, it will be overwritten
  -v, --verbose         When specified, the option will result in the printing of informations on the alignments and the filtering steps (number of
                        ambiguous alignments...etc). Default: OFF
  -q TH_QC, --quality_threshold TH_QC
                        The quality threshold below which alignments will be discarded. The alignment quality index typically ranges between 1 and 41.
                        Default: 30.
  -l TH_LOW, --lowest_size TH_LOW
                        The lowest possible size accepted for DNA fragments. Any pair of reads with an insert size below that value will be discarded.
                        Default: 130.
  -L TH_HIGH, --longest_size TH_HIGH
                        The longest possible size accepted for DNA fragments. Any pair of reads with an insert size above this value will be discarded.
                        Default: 180.
  -d TH_PCR, --duplicate_filter TH_PCR
                        The number of estimated PCR duplicates for accepted for a given genomic location. Default: 1.
  -f FEATURETYPE, --feature_type FEATURETYPE
                        The feature types to be used when generating the count table. The feature type will be matched 3rd column of the GFF file. Default:
                        'Transcript
  -w DOWNSTREAM, --downstream_window DOWNSTREAM
                        The window size used to obtain the counts downstream of the TSS. Default: 1000bp
  -u UPSTREAM, --upstream_window UPSTREAM
                        The window size used to obtain the counts upstream of the TSS. Default: 1500bp

3.4 Example

As an example, let us assume that we want to process an alignment file with the following criteria:

The command line would then be:

python DChIPRep.py -i alignment.sam -a my_gff.gff 
-g chromosome_sizes.txt -o my_count_table.txt -v -q 20 -l 120 
-L 160 -d 2 -f CDS -u 500 -w 500

3.4.1 A note on the count table dimensions

The default number of upstream position is 1500bp, the default number of downstream positions is 1000bp. This results in count tables with 2502 columns in total, corresponding to the basepairs up– and downstream as well as one column for
for the 0 position and one column for the feature IDs.

4 Importing the data into R

In what follows, we assume that we count nucleosomes around transcription start sites (TSS), that is the start of transcripts, which is also the default option in the preprocessing script.

4.1 Importing the output from the Python script

After the data has been preproccesed, we first need an annotation table for our samples that looks like this:

data(exampleSampleTable)
exampleSampleTable
##                               ChIP                          Input upstream downstream  condition
## 1   BY_conf_K4me3_1__ORF_count.txt   BY_conf_WCE_1__ORF_count.txt     1000       1500   K4me3_WT
## 2   BY_conf_K4me3_2__ORF_count.txt   BY_conf_WCE_2__ORF_count.txt     1000       1500   K4me3_WT
## 3   BY_conf_K4me3_3__ORF_count.txt   BY_conf_WCE_3__ORF_count.txt     1000       1500   K4me3_WT
## 4 SET2_conf_K4me3_1__ORF_count.txt SET2_conf_WCE_1__ORF_count.txt     1000       1500 K4me3_SET2
## 5 SET2_conf_K4me3_2__ORF_count.txt SET2_conf_WCE_2__ORF_count.txt     1000       1500 K4me3_SET2
## 6 SET2_conf_K4me3_3__ORF_count.txt SET2_conf_WCE_3__ORF_count.txt     1000       1500 K4me3_SET2
##       sampleID
## 1   K4me3_WT_1
## 2   K4me3_WT_2
## 3   K4me3_WT_3
## 4 K4me3_SET2_1
## 5 K4me3_SET2_2
## 6 K4me3_SET2_3

It gives the names for the input count files in the columns ChIP and Input respectively and the number of base pairs used for analysis upstream and downstream of the TSS in the columns upstream and downstream. Note that the number of upstream and downstream positions needs to be the same for all samples.

The input files must be tab separated files with genomic features in the rows and the positions around the TSS in the columns. In addition to the position columns, a column containing a feature ID must be present.

As mentioned above, the tables have upstream + downstream + 2 columns in total, the two extra columns correspond to the center of the profile (e.g. a TSS) and a column containing the feature IDs.

The sampleID column contains unique sample IDs.

We can then import the data as follows using the function importData which needs the sample annotation table, a data.frame and the directory where the raw data files are stored. We use the down–sampled raw data that come with the package to illustrate the function use.

By default the feature ID column is assumed to have the name name, however this can specified in the call to importData via the ID argument.

directory <- file.path(system.file("extdata", package="DChIPRep"))
importedData <- importData(exampleSampleTable, directory)

The imported data is a DChIPRepResults object that contains the data as a DESeqDataSet. It can be accessed via the function DESeq2Data. The DESeqDataSet also contains normalization factors that are equal to the counts from the chromatin inputs, after being multiplied by the coverage ratio between the input and the ChIP data sets.

importedData
## Summary of the DESeqDataSet:
## ============================
## class: DESeqDataSet 
## dim: 2501 6 
## metadata(0):
## assays(2): counts normalizationFactors
## rownames(2501): Pos_-1000 Pos_-999 ... Pos_1499 Pos_1500
## rowRanges metadata column names(0):
## colnames(6): 1 2 ... 5 6
## colData names(6): ChIP Input ... condition sampleID
## 
## 
## 
## Results:
## ============================
## No results available yet, call runTesting first
DESeq2Data(importedData) 
## class: DESeqDataSet 
## dim: 2501 6 
## metadata(0):
## assays(2): counts normalizationFactors
## rownames(2501): Pos_-1000 Pos_-999 ... Pos_1499 Pos_1500
## rowRanges metadata column names(0):
## colnames(6): 1 2 ... 5 6
## colData names(6): ChIP Input ... condition sampleID
head(normalizationFactors(DESeq2Data(importedData)))
##               1      2      3      4     5      6
## Pos_-1000 3.948 0.8711 2.8502 0.3659 2.535 0.2304
## Pos_-999  5.265 3.0488 1.0688 0.2091 1.268 0.1536
## Pos_-998  1.316 0.8711 0.7126 0.1045 1.268 0.3072
## Pos_-997  7.897 0.4355 1.7814 0.2614 1.690 0.2304
## Pos_-996  2.632 0.8711 3.2065 0.1045 1.268 0.6144
## Pos_-995  2.632 2.1777 2.8502 0.2091 1.690 0.2304

4.2 Importing count matrices

If your data is already available within R, you can import it directly via importDataFromMatrices from an input and a ChIP Matrix. The rows of the matrices should correspond to the positions and the columns to the samples. You furthermore need a sample table as described above, however the columns Input and ChIP are not needed.

If you have data that is still on the level of the individual features, you can use the helper function summarizeCountsPerPosition to summarize the counts for individual features per position.

The package comes with example data sets that correspond to the example sample table shown above.

We first show 10 random rows from the two data matrices and then use the function importDataFromMatrices to import them to a DChIPRepResults object.

As you can see the rows should be named according to the positions up– and downstream of the TSS that they contain and the columns should be named after the samples.

data(exampleInputData)
data(exampleChipData)

exampleSampleTable
##                               ChIP                          Input upstream downstream  condition
## 1   BY_conf_K4me3_1__ORF_count.txt   BY_conf_WCE_1__ORF_count.txt     1000       1500   K4me3_WT
## 2   BY_conf_K4me3_2__ORF_count.txt   BY_conf_WCE_2__ORF_count.txt     1000       1500   K4me3_WT
## 3   BY_conf_K4me3_3__ORF_count.txt   BY_conf_WCE_3__ORF_count.txt     1000       1500   K4me3_WT
## 4 SET2_conf_K4me3_1__ORF_count.txt SET2_conf_WCE_1__ORF_count.txt     1000       1500 K4me3_SET2
## 5 SET2_conf_K4me3_2__ORF_count.txt SET2_conf_WCE_2__ORF_count.txt     1000       1500 K4me3_SET2
## 6 SET2_conf_K4me3_3__ORF_count.txt SET2_conf_WCE_3__ORF_count.txt     1000       1500 K4me3_SET2
##       sampleID
## 1   K4me3_WT_1
## 2   K4me3_WT_2
## 3   K4me3_WT_3
## 4 K4me3_SET2_1
## 5 K4me3_SET2_2
## 6 K4me3_SET2_3
exampleInputData[sample(nrow(exampleInputData), 10), ]
##       K4me3_WT_1 K4me3_WT_2 K4me3_WT_3 K4me3_SET2_1 K4me3_SET2_2 K4me3_SET2_3
## X.22         436        709        975          499          504          488
## X1372       1392       2131       2865         1782         1748         1774
## X.534       1536       2172       3173         1789         1838         1895
## X.577       1678       2292       3302         1929         2083         2164
## X201        2668       3626       5272         3123         3195         3409
## X1275       1502       2083       2856         1791         1735         1854
## X358        2186       2996       4184         2656         2639         2949
## X1299       1550       2276       3116         1861         1805         1861
## X911        1299       1890       2682         1705         1712         1791
## X1454       1568       2298       3222         1891         1937         1919
exampleChipData[sample(nrow(exampleChipData), 10), ]
##       K4me3_WT_1 K4me3_WT_2 K4me3_WT_3 K4me3_SET2_1 K4me3_SET2_2 K4me3_SET2_3
## X.339       3379       1612       1843          149         1191          241
## X.571       5109       2452       2818          237         1898          383
## X1448       3725       1615       1887          187         1461          269
## X.3         2745       1231       1438          108          777          169
## X393       10419       4926       5676          449         3487          667
## X778        3481       1513       1754          161         1282          269
## X.890       4846       2150       2516          226         1781          380
## X536        9808       4516       5108          387         3174          612
## X297        2794       1233       1404          122         1006          213
## X.660       4139       1893       2225          196         1478          304
imDataFromMatrices <- importDataFromMatrices(inputData = exampleInputData, 
                                              chipData = exampleChipData, 
                                              sampleTable = exampleSampleTable)

The imported data is again a DChIPRepResults object that contains the data as a DESeqDataSet.

5 Perform the tests

After the data import, we are ready to perform the tests for differential enrichment. The tests are performed position–wise and wrap DESeq2 and fdrtool. Briefly the DChIPRep testing workflow is as follows:

  1. The function estimateDispersions from DESeq2 is called and the dispersions are estimated.

  2. Then the position–wise tests to compare the experimental conditions are performed. A minimum fold change is used for the null hypothesis, the default value used is 0.05 on a log2 scale.

A possible strategy to infer this threshold from the data is to look a the average fold–change between technical replicates.

  1. The p–values are then passed to fdrtool and the local FDR values are computed.
imDataFromMatrices  <- runTesting(imDataFromMatrices, plotFDR = TRUE)
## Step 1... determine cutoff point
## Step 2... estimate parameters of null distribution and eta0
## Step 3... compute p-values and estimate empirical PDF/CDF
## Step 4... compute q-values and local fdr
## Step 5... prepare for plotting

The results can now be accessed via the function resultsDChIPRep.

res <- resultsDChIPRep(imDataFromMatrices)
head(res)
##           baseMean log2FoldChange   lfcSE     stat pvalue padj lfdr
## Pos_-1000   1013.4       -0.03018 0.05040  0.00000 1.0000 1.00    1
## Pos_-999     978.5       -0.03842 0.05181  0.00000 1.0000 1.00    1
## Pos_-998     996.3        0.05452 0.05073  0.08903 0.9291 1.00    1
## Pos_-997    1056.4       -0.04688 0.04821  0.00000 1.0000 1.00    1
## Pos_-996    1011.3       -0.11971 0.04906 -1.42095 0.1553 0.74    1
## Pos_-995    1005.2       -0.06855 0.05002 -0.37082 0.7108 1.00    1
table( res$lfdr < 0.2)
## 
## FALSE  TRUE 
##  2407    94

At an lfdr of 0.2 we identify 94 significant positions.

6 Plots implemented in the package

6.1 Plot the significant positions

We can first of all plot the TSS profiles by coloring the individual points by significance.

Points corresponding to significant positions are colored black in both of the conditions. The replicate–samples are sumarized by using a positionwise robust Huber estimator for the mean (Hampel, Hennig, and Ronchetti 2011).

The function returns the plot as a ggplot2 object that can be modified afterwards.

sigPlot <- plotSignificance(imDataFromMatrices)
sigPlot

This plot is similar to Figure S17B of (Chabbert et al. 2015). We see an enrichment for significant position near the end of the downstream window considered.

6.2 Produce TSS plots

We can produce the typical, smoothed plots of the TSS profiles as well. Here we use again the smoothed Huber estimator for the mean to compute a summary per experimental group.

profilePlot <- plotProfiles(imDataFromMatrices)
profilePlot

This plot is similar to Figure 5B of (Chabbert et al. 2015).

7 Session information

sessionInfo()
## R version 3.2.3 (2015-12-10)
## Platform: x86_64-w64-mingw32/x64 (64-bit)
## Running under: Windows Server 2008 R2 x64 (build 7601) Service Pack 1
## 
## locale:
## [1] LC_COLLATE=C                           LC_CTYPE=English_United States.1252   
## [3] LC_MONETARY=English_United States.1252 LC_NUMERIC=C                          
## [5] LC_TIME=English_United States.1252    
## 
## attached base packages:
## [1] parallel  stats4    stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] DESeq2_1.10.1              RcppArmadillo_0.6.500.4.0  Rcpp_0.12.3               
##  [4] SummarizedExperiment_1.0.2 Biobase_2.30.0             GenomicRanges_1.22.4      
##  [7] GenomeInfoDb_1.6.3         IRanges_2.4.7              S4Vectors_0.8.11          
## [10] BiocGenerics_0.16.1        ggplot2_2.0.0              DChIPRep_1.0.4            
## [13] rmarkdown_0.9.2            knitr_1.12.3               BiocStyle_1.8.0           
## 
## loaded via a namespace (and not attached):
##  [1] locfit_1.5-9.1       lattice_0.20-33      tidyr_0.4.1          assertthat_0.1      
##  [5] digest_0.6.9         R6_2.1.2             plyr_1.8.3           futile.options_1.0.0
##  [9] acepack_1.3-3.3      RSQLite_1.0.0        evaluate_0.8         zlibbioc_1.16.0     
## [13] lazyeval_0.1.10      annotate_1.48.0      Matrix_1.2-3         rpart_4.1-10        
## [17] labeling_0.3         splines_3.2.3        BiocParallel_1.4.3   geneplotter_1.48.0  
## [21] stringr_1.0.0        foreign_0.8-66       munsell_0.4.2        mgcv_1.8-11         
## [25] htmltools_0.3        nnet_7.3-12          gridExtra_2.0.0      Hmisc_3.17-1        
## [29] codetools_0.2-14     XML_3.98-1.3         dplyr_0.4.3          MASS_7.3-45         
## [33] grid_3.2.3           nlme_3.1-124         xtable_1.8-2         gtable_0.1.2        
## [37] DBI_0.3.1            magrittr_1.5         formatR_1.2.1        scales_0.3.0        
## [41] stringi_1.0-1        XVector_0.10.0       reshape2_1.4.1       genefilter_1.52.1   
## [45] fdrtool_1.2.15       smoothmest_0.1-2     latticeExtra_0.6-26  futile.logger_1.4.1 
## [49] Formula_1.2-1        lambda.r_1.1.7       RColorBrewer_1.1-2   tools_3.2.3         
## [53] survival_2.38-3      yaml_2.1.13          AnnotationDbi_1.32.3 colorspace_1.2-6    
## [57] cluster_2.0.3

References

Chabbert, C. D., S. H. Adjalley, B. Klaus, E. S. Fritsch, I. Gupta, V. Pelechano, and L. M. Steinmetz. 2015. “A High-Throughput ChIP-Seq for Large-Scale Chromatin Studies.” Molecular Systems Biology 11 (1). EMBO: 777–77. doi:10.15252/msb.20145776.

Hampel, Frank, Christian Hennig, and Elvezio Ronchetti. 2011. “A Smoothing Principle for the Huber and Other Location M-Estimators.” Computational Statistics & Data Analysis 55 (1). Elsevier BV: 324–37. doi:10.1016/j.csda.2010.05.001.