e– title: “Package Vignette for Genomic Interactions: ChIA-PET data” output: pdf_document: default html_document: keep_md: TRUE —
ChIA-PET
Chromatin interaction analysis with paired-end tag sequencing (ChIA-PET) is a recent method to study protein-mediated interactions at a genome-wide scale. Like most techniques for studying chromatin interaction it is based on chromosome conformation capture technology. Unlike 3C, 4C and 5C, however, it can detect interactions genome-wide, and includes a ChIP step to purify interactions involving a protein of interest.
The raw data from ChIA-PET is in the form of paired-end reads attached to one of two linker sequences. Reads with chimeric linkers are removed, and the data is aligned to the reference genome. The ChIA-PET tool can then be used to find pairs of regions (“anchors”) which have a significant number of reads mapping between them and therefore represent biologically meaningful chromatin interactions in the sample.
Imports
First we need to load the GenomicInteractions package, and the mm9 reference genome:
library(GenomicInteractions)## Loading required package: InteractionSet## Loading required package: GenomicRanges## Loading required package: stats4## Loading required package: BiocGenerics## Loading required package: parallel##
## Attaching package: 'BiocGenerics'## The following objects are masked from 'package:parallel':
##
## clusterApply, clusterApplyLB, clusterCall, clusterEvalQ,
## clusterExport, clusterMap, parApply, parCapply, parLapply,
## parLapplyLB, parRapply, parSapply, parSapplyLB## The following objects are masked from 'package:stats':
##
## IQR, mad, sd, var, xtabs## The following objects are masked from 'package:base':
##
## Filter, Find, Map, Position, Reduce, anyDuplicated, append,
## as.data.frame, cbind, colMeans, colSums, colnames, do.call,
## duplicated, eval, evalq, get, grep, grepl, intersect,
## is.unsorted, lapply, lengths, mapply, match, mget, order,
## paste, pmax, pmax.int, pmin, pmin.int, rank, rbind, rowMeans,
## rowSums, rownames, sapply, setdiff, sort, table, tapply,
## union, unique, unsplit, which, which.max, which.min## Loading required package: S4Vectors##
## Attaching package: 'S4Vectors'## The following object is masked from 'package:base':
##
## expand.grid## Loading required package: IRanges## Loading required package: GenomeInfoDb## Loading required package: SummarizedExperiment## Loading required package: Biobase## Welcome to Bioconductor
##
## Vignettes contain introductory material; view with
## 'browseVignettes()'. To cite Bioconductor, see
## 'citation("Biobase")', and for packages 'citation("pkgname")'.## Loading required package: DelayedArray## Loading required package: matrixStats##
## Attaching package: 'matrixStats'## The following objects are masked from 'package:Biobase':
##
## anyMissing, rowMedians##
## Attaching package: 'DelayedArray'## The following objects are masked from 'package:matrixStats':
##
## colMaxs, colMins, colRanges, rowMaxs, rowMins, rowRanges## The following object is masked from 'package:base':
##
## applylibrary(InteractionSet)
library(GenomicRanges)Data
We can then read in our data directly from the output of the ChIA-PET tool. At this stage we can also provide information about the cell type and a description tag for the experiment. The data is taken from Li et al., 2012, published in Cell. They have used antibodies against the initiation form of Pol II, which you would expect to find at active promoters, and we are looking at data from the K562 myelogenous leukemia cell line. The data should therefore give us an insight into the processes which regulate genes that are being actively transcribed.
chiapet.data = system.file("extdata/k562.rep1.cluster.pet3+.txt",
package="GenomicInteractions")
k562.rep1 = makeGenomicInteractionsFromFile(chiapet.data,
type="chiapet.tool",
experiment_name="k562",
description="k562 pol2 8wg16")This loads the data into a GenomicInteractions object, which consists of two linked GenomicRanges objects containing the anchors in each interaction, as well as the p-value, FDR and the number of reads supporting each interaction.
GenomicInteractions Objects
The metadata we have added can easily be accesed, and edited:
name(k562.rep1)## [1] "k562"description(k562.rep1) = "PolII-8wg16 Chia-PET for K562"As can the data from the ChIA-PET experiment:
head(interactionCounts(k562.rep1))## [1] 3 562 3 3 3 3head((k562.rep1)$fdr)## [1] 1.25703e-10 0.00000e+00 1.17148e-06 4.86859e-08 2.76777e-08 3.97019e-08hist(-log10(k562.rep1$p.value))
The two linked GRanges objects can be returned, but not altered in-place:
anchorOne(k562.rep1)## GRanges object with 64565 ranges and 0 metadata columns:
## seqnames ranges strand
## <Rle> <IRanges> <Rle>
## [1] chr1 [569922, 571422] *
## [2] chr1 [832761, 905482] *
## [3] chr1 [839092, 842325] *
## [4] chr1 [839393, 841792] *
## [5] chr1 [852731, 855234] *
## ... ... ... ...
## [64561] chrX [154432946, 154435728] *
## [64562] chrX [154436728, 154439876] *
## [64563] chrX [154439789, 154442306] *
## [64564] chrX [154459648, 154462031] *
## [64565] chrX [154839050, 154843949] *
## -------
## seqinfo: 25 sequences from an unspecified genome; no seqlengthsanchorTwo(k562.rep1)## GRanges object with 64565 ranges and 0 metadata columns:
## seqnames ranges strand
## <Rle> <IRanges> <Rle>
## [1] chrM [ 8342, 10675] *
## [2] chr1 [838470, 920603] *
## [3] chr1 [935528, 939051] *
## [4] chr1 [955081, 956755] *
## [5] chr1 [933685, 937006] *
## ... ... ... ...
## [64561] chrX [154442294, 154446983] *
## [64562] chrX [154442540, 154445105] *
## [64563] chrX [154448371, 154451728] *
## [64564] chrX [154469339, 154471852] *
## [64565] chrX [154843728, 154848393] *
## -------
## seqinfo: 25 sequences from an unspecified genome; no seqlengthsGenomicInteractions objects can easily handle interactions detected between chromosomes, known as trans-chromosomal interactions, since the anchors can be at any point along the genome. is.trans returns a logical vector; likewise is.cis is the opposite of this function.
sprintf("Percentage of trans-chromosomal interactions %.2f",
100*sum(is.trans(k562.rep1))/length(k562.rep1))## [1] "Percentage of trans-chromosomal interactions 1.00"The length of each interaction is not stored as metadata, but we can calculate the distance of each interaction using either the inner edge, outer edge or midpoints of the anchors. This is undefined for inter-chromosomal interactions, so NA is returned, so it is important to exclude these interactions from some analyses.
head(calculateDistances(k562.rep1, method="midpoint"))## [1] NA 10415 96581 115325 81363 79098GenomicRanges objects can be subsetted by either integer or logical vectors like most R objects, and also BioConductor Rle objects.
k562.rep1[1:10] # first interactions in the dataset## GenomicInteractions object with 10 interactions and 3 metadata columns:
## seqnames1 ranges1 seqnames2 ranges2 |
## <Rle> <IRanges> <Rle> <IRanges> |
## [1] chr1 [569922, 571422] --- chrM [ 8342, 10675] |
## [2] chr1 [832761, 905482] --- chr1 [838470, 920603] |
## [3] chr1 [839092, 842325] --- chr1 [935528, 939051] |
## [4] chr1 [839393, 841792] --- chr1 [955081, 956755] |
## [5] chr1 [852731, 855234] --- chr1 [933685, 937006] |
## [6] chr1 [855856, 858861] --- chr1 [935669, 937245] |
## [7] chr1 [874165, 879175] --- chr1 [933340, 938306] |
## [8] chr1 [874190, 877867] --- chr1 [955674, 959630] |
## [9] chr1 [889676, 896594] --- chr1 [933897, 938982] |
## [10] chr1 [898753, 907581] --- chr1 [931133, 939571] |
## counts p.value fdr
## <integer> <numeric> <numeric>
## [1] 3 1.6214e-12 1.25703e-10
## [2] 562 0 0
## [3] 3 4.21364e-08 1.17148e-06
## [4] 3 1.45938e-09 4.86859e-08
## [5] 3 7.85539e-10 2.76777e-08
## [6] 3 1.16802e-09 3.97019e-08
## [7] 10 1.23139e-25 3.58932e-23
## [8] 5 6.63691e-15 6.98795e-13
## [9] 13 4.91311e-36 2.33753e-33
## [10] 19 0 0
## -------
## regions: 129090 ranges and 0 metadata columns
## seqinfo: 25 sequences from an unspecified genome; no seqlengthsk562.rep1[sample(length(k562.rep1), 100)] # 100 interactions subsample## GenomicInteractions object with 100 interactions and 3 metadata columns:
## seqnames1 ranges1 seqnames2
## <Rle> <IRanges> <Rle>
## [1] chrX [133674612, 133678396] --- chrX
## [2] chr21 [ 30375160, 30377049] --- chr21
## [3] chr8 [144459016, 144463205] --- chr8
## [4] chr5 [145501690, 145508213] --- chr5
## [5] chr10 [ 32323738, 32335123] --- chr10
## ... ... ... ... ...
## [96] chr13 [100606804, 100630989] --- chr13
## [97] chr12 [ 22201346, 22204337] --- chr12
## [98] chr13 [114765941, 114770164] --- chr13
## [99] chr19 [ 5955506, 5958490] --- chr19
## [100] chr17 [ 40686431, 40702265] --- chr17
## ranges2 | counts p.value fdr
## <IRanges> | <integer> <numeric> <numeric>
## [1] [133678146, 133683756] | 9 3.64628e-37 1.80206e-34
## [2] [ 30405892, 30408661] | 3 8.5231e-12 5.66267e-10
## [3] [144509268, 144513217] | 5 1.56342e-17 2.37031e-15
## [4] [145506471, 145511935] | 7 4.74138e-31 1.83101e-28
## [5] [ 32333013, 32341678] | 17 0 0
## ... ... . ... ... ...
## [96] [100615362, 100644941] | 121 0 0
## [97] [ 22211554, 22214294] | 3 1.28312e-13 1.13691e-11
## [98] [114998593, 115001693] | 3 1.01126e-11 6.52726e-10
## [99] [ 5961777, 5963601] | 3 9.31404e-14 8.37916e-12
## [100] [ 40703518, 40709901] | 29 0 0
## -------
## regions: 129090 ranges and 0 metadata columns
## seqinfo: 25 sequences from an unspecified genome; no seqlengthsk562.cis = k562.rep1[is.cis(k562.rep1)]The length of each interaction is not stored as metadata, but we can calculate the distance of each interaction using either the inner edge, outer edge or midpoints of the anchors. Since this is undefinable for trans-chromosomal interactions it is best to first subset only cis interactions before calling calculateDistances, otherwise NAs will be present in the returned vector.
head(calculateDistances(k562.cis, method="midpoint"))## [1] 10415 96581 115325 81363 79098 59153k562.short = k562.cis[calculateDistances(k562.cis) < 1e6] # subset shorter interactions
hist(calculateDistances(k562.short)) 
We can also subset based on the properties of the linked GRanges objects.
chrom = c("chr17", "chr18")
sub = as.vector(seqnames(anchorOne(k562.rep1)) %in% chrom & seqnames(anchorTwo(k562.rep1)) %in% chrom)
k562.rep1 = k562.rep1[sub]Annotation
Genomic Interaction data is often used to look at the interactions between different elements in the genome, which are believed to have different functional roles. Interactions between promoters and their transcription termination sites, for example, are thought to be a by-product of the transcription process, whereas long-range interactions with enhancers play a role in gene regulation.
Since GenomicInteractions is based on GenomicRanges, it is very easy to interrogate GenomicInteractions objects using GenomicRanges data. In the example, we want to annotate interactions that overlap the promoters, transcription termination sites or the body of any gene. Since this can be a time-consuming and data-heavy process, this example runs the analysis for only chromosomes 17 & 18.
First we need the list of RefSeq transcripts:
library(GenomicFeatures)
hg19.refseq.db <- makeTxDbFromUCSC(genome="hg19", table="refGene")
refseq.genes = genes(hg19.refseq.db)
refseq.transcripts = transcriptsBy(hg19.refseq.db, by="gene")
non_pseudogene = names(refseq.transcripts) %in% unlist(refseq.genes$gene_id)
refseq.transcripts = refseq.transcripts[non_pseudogene] Rather than downloading the whole Refseq database, these are provided for chromosomes 17 & 18:
data("hg19.refseq.transcripts")
refseq.transcripts = hg19.refseq.transcriptsWe can then use functions from GenomicRanges to call promoters and terminators for these transcripts. We have taken promoter regions to be within 2.5kb of an annotated TSS and terminators to be within 1kb of the end of an annotated transcript. Since genes can have multiple transcripts, they can also have multiple promoters/terminators, so these are GRangesList objects, which makes handling these objects slightly more complicated.
refseq.promoters = promoters(refseq.transcripts, upstream=2500, downstream=2500)
# unlist object so "strand" is one vector
refseq.transcripts.ul = unlist(refseq.transcripts)
# terminators can be called as promoters with the strand reversed
strand(refseq.transcripts.ul) = ifelse(strand(refseq.transcripts.ul) == "+", "-", "+")
refseq.terminators.ul = promoters(refseq.transcripts.ul, upstream=1000, downstream=1000)
# change back to original strand
strand(refseq.terminators.ul) = ifelse(strand(refseq.terminators.ul) == "+", "-", "+")
# `relist' maintains the original names and structure of the list
refseq.terminators = relist(refseq.terminators.ul, refseq.transcripts)These can be used to subset a GenomicInteractions object directly from GRanges using the GenomicRanges overlaps methods. findOverlaps called on a GenomicInteractions object will return a list containing Hits objects for both anchors.
We can finds any interactions involving a RefSeq promoter:
subsetByFeatures(k562.rep1, unlist(refseq.promoters))## GenomicInteractions object with 2907 interactions and 3 metadata columns:
## seqnames1 ranges1 seqnames2 ranges2
## <Rle> <IRanges> <Rle> <IRanges>
## [1] chr17 [616579, 621961] --- chr17 [620668, 626263]
## [2] chr17 [632527, 638035] --- chr17 [636589, 641349]
## [3] chr17 [634119, 651606] --- chr17 [642299, 659172]
## [4] chr17 [654892, 657597] --- chr17 [683191, 687275]
## [5] chr17 [656002, 658841] --- chr17 [679595, 682692]
## ... ... ... ... ... ...
## [2903] chr18 [77781151, 77783476] --- chr18 [77792968, 77795855]
## [2904] chr18 [77784590, 77787797] --- chr18 [77792148, 77795822]
## [2905] chr18 [77792093, 77797983] --- chr18 [77797455, 77803127]
## [2906] chr18 [77793365, 77797939] --- chr18 [77864889, 77868321]
## [2907] chr18 [77863992, 77870413] --- chr18 [77868294, 77877151]
## | counts p.value fdr
## | <integer> <numeric> <numeric>
## [1] | 7 2.06358e-32 8.50563e-30
## [2] | 9 2.395e-36 1.1518e-33
## [3] | 55 0 0
## [4] | 6 4.86283e-24 1.27856e-21
## [5] | 3 6.23098e-14 5.72161e-12
## ... . ... ... ...
## [2903] | 3 8.97548e-14 8.09366e-12
## [2904] | 6 4.16341e-26 1.24576e-23
## [2905] | 13 0 0
## [2906] | 8 6.61618e-30 2.41717e-27
## [2907] | 18 0 0
## -------
## regions: 129090 ranges and 0 metadata columns
## seqinfo: 25 sequences from an unspecified genome; no seqlengthsHowever, one of the most powerful features in the GenomicInteractions package is the ability to annotate each anchor with a list of genomic regions and then summarise interactions according to these features. This annotation is implemented as metadata columns for the anchors in the GenomicInteractions object and so is fast, and facilitates more complex analyses.
The order in which we annotate the anchors is important, since each anchor can only have one node.class. The first listed take precedence. Any regions not overlapping ranges in annotation.features will be labelled as distal.
annotation.features = list(promoter=refseq.promoters,
terminator=refseq.terminators,
gene.body=refseq.transcripts)
annotateInteractions(k562.rep1, annotation.features)## Annotating with promoter ...## Annotating with terminator ...## Annotating with gene.body ...annotationFeatures(k562.rep1)## [1] "distal" "gene.body" "promoter" "terminator"We can now find interactions involving promoters using the annotated node.class for each anchor:
p.one = anchorOne(k562.rep1)$node.class == "promoter"
p.two = anchorTwo(k562.rep1)$node.class == "promoter"
k562.rep1[p.one|p.two]## GenomicInteractions object with 2907 interactions and 3 metadata columns:
## seqnames1 ranges1 seqnames2 ranges2
## <Rle> <IRanges> <Rle> <IRanges>
## [1] chr17 [616579, 621961] --- chr17 [620668, 626263]
## [2] chr17 [632527, 638035] --- chr17 [636589, 641349]
## [3] chr17 [634119, 651606] --- chr17 [642299, 659172]
## [4] chr17 [654892, 657597] --- chr17 [683191, 687275]
## [5] chr17 [656002, 658841] --- chr17 [679595, 682692]
## ... ... ... ... ... ...
## [2903] chr18 [77781151, 77783476] --- chr18 [77792968, 77795855]
## [2904] chr18 [77784590, 77787797] --- chr18 [77792148, 77795822]
## [2905] chr18 [77792093, 77797983] --- chr18 [77797455, 77803127]
## [2906] chr18 [77793365, 77797939] --- chr18 [77864889, 77868321]
## [2907] chr18 [77863992, 77870413] --- chr18 [77868294, 77877151]
## | counts p.value fdr
## | <integer> <numeric> <numeric>
## [1] | 7 2.06358e-32 8.50563e-30
## [2] | 9 2.395e-36 1.1518e-33
## [3] | 55 0 0
## [4] | 6 4.86283e-24 1.27856e-21
## [5] | 3 6.23098e-14 5.72161e-12
## ... . ... ... ...
## [2903] | 3 8.97548e-14 8.09366e-12
## [2904] | 6 4.16341e-26 1.24576e-23
## [2905] | 13 0 0
## [2906] | 8 6.61618e-30 2.41717e-27
## [2907] | 18 0 0
## -------
## regions: 129090 ranges and 4 metadata columns
## seqinfo: 25 sequences from an unspecified genome; no seqlengthsThis information can be used to categorise interactions into promoter-distal, promoter-terminator etc. A table of interaction types can be generated with categoriseInteractions:
categoriseInteractions(k562.rep1)## category count
## 1 distal-distal 396
## 2 distal-gene.body 76
## 3 distal-promoter 519
## 4 distal-terminator 101
## 5 gene.body-gene.body 795
## 6 gene.body-promoter 917
## 7 gene.body-terminator 164
## 8 promoter-promoter 1187
## 9 promoter-terminator 284
## 10 terminator-terminator 70Alternatively, we can subset the object based on interaction type:
k562.rep1[isInteractionType(k562.rep1, "terminator", "gene.body")]## GenomicInteractions object with 164 interactions and 3 metadata columns:
## seqnames1 ranges1 seqnames2 ranges2
## <Rle> <IRanges> <Rle> <IRanges>
## [1] chr17 [1471460, 1474306] --- chr17 [1476212, 1479585]
## [2] chr17 [1632603, 1638741] --- chr17 [1636657, 1642967]
## [3] chr17 [3845975, 3849573] --- chr17 [3908008, 3910817]
## [4] chr17 [4055645, 4058706] --- chr17 [4063341, 4068158]
## [5] chr17 [4443889, 4451615] --- chr17 [4446814, 4454998]
## ... ... ... ... ... ...
## [160] chr18 [32873043, 32876330] --- chr18 [32911004, 32914907]
## [161] chr18 [33566139, 33569380] --- chr18 [33571111, 33574360]
## [162] chr18 [33689982, 33692774] --- chr18 [33695495, 33699363]
## [163] chr18 [42638745, 42641928] --- chr18 [42647110, 42649707]
## [164] chr18 [77914125, 77916781] --- chr18 [77919050, 77922761]
## | counts p.value fdr
## | <integer> <numeric> <numeric>
## [1] | 4 1.1438e-18 1.92712e-16
## [2] | 12 0 4.06377e-44
## [3] | 5 1.46527e-19 2.69952e-17
## [4] | 4 4.96495e-19 8.66918e-17
## [5] | 11 7.2293e-42 4.17953e-39
## ... . ... ... ...
## [160] | 4 1.37802e-15 1.62547e-13
## [161] | 3 5.37957e-15 5.74333e-13
## [162] | 4 1.35703e-19 2.5096e-17
## [163] | 3 1.03659e-15 1.24713e-13
## [164] | 3 2.9097e-15 3.24878e-13
## -------
## regions: 129090 ranges and 4 metadata columns
## seqinfo: 25 sequences from an unspecified genome; no seqlengthsThe 3 most common node.class values have short functions defined for convenience (see ?is.pp for a complete list):
k562.rep1[is.pp(k562.rep1)] # promoter-promoter interactions
k562.rep1[is.dd(k562.rep1)] # distal-distal interactions
k562.rep1[is.pt(k562.rep1)] # promoter-terminator interactionsSummary plots of interactions classes can easily be produced to get an overall feel for the data:
plotInteractionAnnotations(k562.rep1, other=5)
viewpoints will only take those interactions with a certain node.class:
plotInteractionAnnotations(k562.rep1, other=5, viewpoints="promoter")
These are also combined in the function plotSummaryStats.
Feature Summaries
The summariseByFeatures allows us to look in more detail at interactions involving a specific set of loci. In this example we use all RefSeq promoters, which we already have loaded in a GRangesList object.
It is however possible to use any dataset which can be represented as a named GRanges object, for example transcription-factor ChIP data, predicted cis-regulatory sites or certain categories of genes.
The categories are generated automatically from the annotated node.class values in the object.
k562.rep1.promoter.annotation = summariseByFeatures(k562.rep1, refseq.promoters,
"promoter", distance.method="midpoint",
annotate.self=TRUE)
colnames(k562.rep1.promoter.annotation)## [1] "Promoter.id"
## [2] "numberOfPromoterInteractions"
## [3] "numberOfPromoterUniqueInteractions"
## [4] "numberOfPromoterInterChromosomalInteractions"
## [5] "numberOfPromoterUniqueInterChromosomalInteractions"
## [6] "numberOfPromoterDistalInteractions"
## [7] "numberOfPromoterGene.bodyInteractions"
## [8] "numberOfPromoterPromoterInteractions"
## [9] "numberOfPromoterTerminatorInteractions"
## [10] "numberOfUniquePromoterDistalInteractions"
## [11] "numberOfUniquePromoterGene.bodyInteractions"
## [12] "numberOfUniquePromoterPromoterInteractions"
## [13] "numberOfUniquePromoterTerminatorInteractions"
## [14] "PromoterDistanceMedian"
## [15] "PromoterDistanceMean"
## [16] "PromoterDistanceMinimum"
## [17] "PromoterDistanceMaximum"
## [18] "PromoterDistanceWeightedMedian"
## [19] "numberOfSelfPromoterGene.bodyInteractions"
## [20] "numberOfSelfPromoterPromoterInteractions"
## [21] "numberOfSelfPromoterTerminatorInteractions"
## [22] "numberOfSelfUniquePromoterGene.bodyInteractions"
## [23] "numberOfSelfUniquePromoterPromoterInteractions"
## [24] "numberOfSelfUniquePromoterTerminatorInteractions"This allows us to very quickly generate summaries of the data and provides a quick method to isolate genes of interest. In this case we produce lists of RefSeq IDs, which can easily be converted to EntrezIDs or gene symbols through existing BioConductor packages (in this case org.Hs.eg.db provides bimaps between common human genome annotations).
Which promoters have the strongest Promoter-Promoter interactions based on PET-counts?
i = order(k562.rep1.promoter.annotation$numberOfPromoterPromoterInteractions,
decreasing=TRUE)[1:10]
k562.rep1.promoter.annotation[i,"Promoter.id"]## [1] "100506779" "9256" "406934" "54894" "100616220"
## [6] "6827" "56155" "5889" "5034" "396"Which promoters are contacting the largest number of distal elements?
i = order(k562.rep1.promoter.annotation$numberOfUniquePromoterDistalInteractions,
decreasing=TRUE)[1:10]
k562.rep1.promoter.annotation[i,"Promoter.id"]## [1] "10140" "400604" "7050" "100130581" "100616277"
## [6] "26118" "100874261" "101927666" "140735" "5366"What percentage of promoters are in contact with transcription termination sites?
total = sum(k562.rep1.promoter.annotation$numberOfPromoterTerminatorInteractions > 0)
sprintf("%.2f%% of promoters have P-T interactions", 100*total/nrow(k562.rep1.promoter.annotation))## [1] "16.43% of promoters have P-T interactions"References
Li, Guoliang, et al. “Software ChIA-PET tool for comprehensive chromatin interaction analysis with paired-end tag sequencing.” Genome Biol 11 (2010): R22.
Li, Guoliang, et al. “Extensive promoter-centered chromatin interactions provide a topological basis for transcription regulation.” Cell 148.1 (2012): 84-98