The COMRADES experimental protocol for the prediction of RNA structure
in vivo was first published in 2018 (Ziv et al., 2019) where they
predicted the structure of the Zika virus. The protocol has subsequently
been use to predict the structure of SARS-CoV-2 (Ziv et al., 2020). Have
a look to get an understanding of the protocol:
COMRADES determines in vivo RNA structures and interactions. (2018). Omer Ziv, Marta Gabryelska, Aaron Lun, Luca Gebert. Jessica Sheu-Gruttadauria and Luke Meredith, Zhong-Yu Liu, Chun Kit Kwok, Cheng-Feng Qin, Ian MacRae, Ian Goodfellow , John Marioni, Grzegorz Kudla, Eric Miska. Nature Methods. Volume 15. https://doi.org/10.1038/s41592-018-0121-0
The Short- and Long-Range RNA-RNA Interactome of SARS-CoV-2. (2020). Omer Ziv, Jonathan Price, Lyudmila Shalamova, Tsveta Kamenova, Ian Goodfellow, Friedemann Weber, Eric A. Miska. Molecular Cell, Volume 80 https://doi.org/10.1016/j.molcel.2020.11.004
After sequencing, short reads are produced similar to a spliced / chimeric RNA read but where one half of the read corresponds to one half of a structural RNA duplex and the other half of the reads corresponds to the other half of the structural RNA duplex. This package has been designed to analyse this data. The short reads need to be prepared in a specific way to be inputted into this package.
There are other types of crosslinking data!
PARIS
Fastq files produced from the comrades experiment can be processed for
input into comradesOO using the Nextflow pre-processing pipeline, to get
more information visit here.. The pipeline takes the reads through
trimming alignment, QC and the production of the files necessary for
input to comradesOO. Crosslinking experiments often have different
library preparation protocols therefore it is not necessary to follow
the prescribed pre-processing pipeline. The only requirement is that the
input files for comradesOO have the correct format detailed below.
The main output files are the files entitled X_gapped.txt.
These are the input files for comradesOO. The columns of the output
files are as follows:
The main input files for comrades-OO is a tab delimited text file
containing the reads and mapping location on the transcriptome. This can
be manually created if your library preparation protocol does not suit
the pre-processing pipeline although the easiest way to obtain these
files is to use the nextflow pipeline detailed above. There is test data
that ships with the package, this contains data for the 18S rRNA and
it’s interactions with the 28S rRNA. However, full data-sets already
published can be found here:Un-enriched
rRNA dataset.
Pre-requisites:
There is a development version available on github (https://github.com/JLP-BioInf/comradesOO). Issue reporting and collaboration welcome.
# Load the comrades-OO Library
#install.packages("comradesOO")
library(comradesOO)
The package relies on functions from these packages:
# Here are the other libraries on which comradesOO relies
library(seqinr)
library(GenomicRanges)
library(ggplot2)
library(reshape2)
library(MASS)
library(ggplot2)
library(doParallel)
library(igraph)
library(R4RNA)
library(RColorBrewer)
library(heatmap3)
library(mixtools)
library(TopDom)
library(tidyverse)
library(RRNA)
library(ggrepel)
The metadata table has 4 columns and the column names are specific and case-sensitive.
# Set up the sample table
= c(system.file("extdata",
sampleTableRow1 "s1.txt",
package="comradesOO"), "s", "1", "s1")
= c(system.file("extdata",
sampleTableRow2 "c1.txt",
package="comradesOO"), "c", "1", "c1")
= rbind.data.frame(sampleTableRow1, sampleTableRow2)
sampleTable2
# add the column names
colnames(sampleTable2) = c("file", "group", "sample", "sampleName")
sampleTable2
The name of the RNA to analyse, this must be as it appears in the input files.
= c("ENSG000000XXXXX_NR003286-2_RN18S1_rRNA") rna
Fasta sequence(s) of the RNA(s) of interest, taken from the transcriptome reference fasta used for mapping. Load in using the read.fasta function from seqinr.
<- system.file("extdata",
path18SFata "18S.fasta",
package="comradesOO")
= list()
rnaRefs = read.fasta(path18SFata) rnaRefs[[rna]]
This is optional but you can provide a table of interactions for the RNA to compare against. This can be useful when comparing different samples or to another predicted structure for the same RNA. The table should be a tsv with to columns (i and j) each row shows an interaction between nucleotide i and j for comparison.
<- system.file("extdata",
path18SFata "ribovision18S.txt",
package="comradesOO")
= read.table(path18SFata,
known18S header = F)
If you also have reactivities from a chemical probing experiment they can be included here as a 1 column table with one value for each position in the transcript. This feature is not ready.
<- system.file("extdata",
pathShape "reactivities.txt",
package="comradesOO")
= read.table(pathShape,
shape header = F)
The package has 3 main processes; clustering, cluster trimming and folding. The next sections take you through the usage of each of these main stages.
The instance of the comradesDataSet object that is created stores the information from the experiment including raw and processed data for the dataset.
The slots for processed and unprocessed data keep the data from each
stage of the analysis, this allows the user to quickly access any part
of the results. Checking the status of the object will allow you to see
which stages of the analysis are present for each of the
attributes.
# load the object
= comradesDataSet(rnas = rna,
cds sampleTable = sampleTable2)
#> ********************************************
#> ***** COMRADES-OO ******
#> ********************************************
#> *****-------*******************-------******
#> ***** Reading SampleTable ******
#> ***** Detected 2 Samples ******
#> ***** detected group c:: 2 *****
#> ***** detected group s:: 1 *****
#> **** Sample Names: s1 c1 **** **** Sample Names: s1 c1 **** **** Sample Names: s1 c1 **** **** Sample Names: s1 c1 ****
#> ***** Reading Hyb Files *****
#> ***** Getting RNAs of Interest ******
#> ***** RNA of interest + Host RNA *****
#> ***** RNA of interest Alone *****
#> ***** Making Matrices ******
#> ***** RNA Size: 1870 *****
#> ***** Creating object *****
#> *****-------*******************-------******
#> ********************************************
#> ********************************************
You can check on major parts of the object and return slots and other information using the accessor methods
# Check status of instance
cds#> comradesDataSet Object
#> RNAs Analysed - - ENSG000000XXXXX_NR003286-2_RN18S1_rRNA
#> Samples Analysed - - s1 c1
#> hybFiles - Raw data - original host noHost
#> matrixList - Contact Matricies - noHost original
#> clusterTableList - Cluster positions -
#> clusterGrangesList - Granges of clusters -
#> interactionTable - Predictied interaction - 0
#> viennaStructures - Predicted Structures - 0
# Returns the size of the RNA
rnaSize(cds)
#> [1] 1870
# Returns the sample table
sampleTable(cds)
# Returns indexes of the samples in the control and not control groups
group(cds)
#> $c
#> [1] 2
#>
#> $s
#> [1] 1
# Get the sample names of the instance
sampleNames(cds)
#> [1] "s1" "c1"
It is more recommended to use getData for this purpose but sometime is is useful to grab all data in the hybFiles slot which contains all raw and processed data in the original input format from each analysis stage that has been performed.
# Return the hybFiles slot
hybFiles(cds)
It is more recommended to use getData for this purpose but sometime is is useful to grab all data in the matrixList slot which contains contact matrices from each analysis stage that has been performed.
# Return the matrixList slot
matrixList(cds)
Get data is more generic method for retrieving data from the object and returns a list, the number of entries in the list is number of samples in the dataset and the list contain entries of the data type and analysis stage you select.
= getData(x = cds, # The object
data data = "hybFiles", # The Type of data to return
type = "original")[[1]] # The stage of the analysis for the return data
head(data)
The first step is to assess the species of RNA present in the dataset, the instance will probably contain inter-RNA interactions and intra-RNA interactions for many different RNAs. A number of tables showing the different RNAs / interactions and the ammount of reads assigned to each can be returned with the following methods:
# Returns the RNAs with highest number of assigned reads
# regardless of wether it is an Inter or Intra - RNA interaction.
topTranscripts(cds, # The comradesDataSet instance
2) # The number of entried to return
# Returns the RNAs that interact with the RNA of interest
topInteracters(cds, # The comradesDataSet instance
1) # The number of entries to return
# Returns the Interacions with the highest number of assigned reads
topInteractions(cds, # The comradesDataSet instance
2) # The number of entries to return
= featureInfo(cds) # The comradesDataSet instance features
# Counts for features at the transcript level
$transcript features
# Counts for features at the family level (last field with "_" delimited IDs)
$family features
In the COMRADES data, crosslinking and fragmentation leads to the production of redundant structural information, where the same in vivo structure from different RNA molecules produces slightly different RNA fragments. Clustering of these duplexes that originate from the same place in the reference transcript reduces computational time and allows trimming of these clusters to improve the folding prediction. To allow clustering, gapped alignments can be described by the transcript coordinates of the left (L) and right (R) side of the reads and by the nucleotides between L and R (g). Reads with similar or identical g values are likely to originate from the same structure of different molecules. In COMRADES-OO, an adjacency matrix is created for all chimeric reads based on the nucleotide difference between their g values (Deltagap). This results in Deltagap = 0 for identically overlapping gaps and increasing Deltagap values for gapped reads with less overlap:
For short range interactions ( g <= 10 nt ) the weights are calculated such that the highest weights are given to exactly overlapping gapped alignments and a weight of 0 is assigned to alignments that do not overlap.
Long range interactions (g >10) are clustered separately and their weights are calculated as follows and edges with weights lower that 0 are set to 0. Meaning that gaps that do not overlap by at least 15 nucleotides are considered in different clusters.
From these weights the network can be defined for short- and long-range interaction as: G = (V, E). To identify clusters within the graph (subgraphs) the graph is clustered using random walks with the cluster_waltrap function (steps = 2) from the iGraph packageå, there is an option for users to remove clusters with less than a specified amount of reads. These clusters often contain a small number of longer L or R sequences due to the random fragmentation in the COMRADES protocol.
# Cluster the reads
= clusterComrades(cds = cds, # The comradesDataSet instance
clusteredCds cores = 1, # The number of cores
stepCount = 2, # The number of steps in the random walk
clusterCutoff = 1) # The minimum number of reads for a cluster to be considered
#> ********************************************
#> **** ENSG000000XXXXX_NR003286-2_RN18S1_rRNA *****
#> **** 1870 nt ****
#> **** Assessing Long Range ****
#> **** Sampling Long Range ****
#> **** Assessing Short Range ****
#> **** Sampling short Range ****
#> ***** done s1 1 / 1 *****
#> ***** done c1 1 / 1 *****
#> ***** Creating object *****
#> ********************************************
# Check status of instance
clusteredCds#> comradesDataSet Object
#> RNAs Analysed - - ENSG000000XXXXX_NR003286-2_RN18S1_rRNA
#> Samples Analysed - - s1 c1
#> hybFiles - Raw data - original host noHost
#> matrixList - Contact Matricies - noHost original originalClusters
#> clusterTableList - Cluster positions - original
#> clusterGrangesList - Granges of clusters - original
#> interactionTable - Predictied interaction - 0
#> viennaStructures - Predicted Structures - 0
# Returns the number of clusters in each sample
clusterNumbers(clusteredCds)
# Returns the number reads in clusters
readNumbers( clusteredCds)
The cluster tables contain coordinates of the clusters in data.frame format. Each cluster has a unique ID and size.x corrasponds to the number of reads assigned to that cluster or supercluster. ls, le, rs and le give the coordinates of the interaction.
getData(clusteredCds, # The object
"clusterTableList", # The Type of data to return
"original")[[1]] # The stage of the analysis for the return data
You can also extract a GRanges object of the individual reads and their cluster membership:
getData(clusteredCds, # The object
"clusterGrangesList", # The Type of data to return
"original")[[1]] # The stage of the analysis for the return data
Given the assumption that the reads within each cluster likely originate from the same structure in different molecules these clusters can be trimmed to contain the regions from L and R that have the most evidence the clustering and trimming is achieved with the clusterComrades and trimClusters methods.
# Trim the Clusters
= trimClusters(clusteredCds = clusteredCds, # The comradesDataSet instance
trimmedClusters trimFactor = 1, # The cutoff for cluster trimming (see above)
clusterCutoff = 0) # The minimum number of reads for a cluster to be considered
#> ********************************************
#> ****** Trimming Clusters ******
#> ****** Saving ******
#> ****** Saving mat list ******
#> ****** Saving table list ******
#> ****** Saving End ******
#> ****** Saving mat list End ******
#> ****** Saving granges list ******
#> ****** Saving table list End ******
#> ********************************************
# Check status of instance
trimmedClusters#> comradesDataSet Object
#> RNAs Analysed - - ENSG000000XXXXX_NR003286-2_RN18S1_rRNA
#> Samples Analysed - - s1 c1
#> hybFiles - Raw data - original host noHost
#> matrixList - Contact Matricies - noHost original originalClusters superClusters trimmedClusters
#> clusterTableList - Cluster positions - original superClusters trimmedClusters
#> clusterGrangesList - Granges of clusters - original superClusters trimmedClusters
#> interactionTable - Predictied interaction - 0
#> viennaStructures - Predicted Structures - 0
# Returns the number of clusters in each sample
clusterNumbers(trimmedClusters)
# Returns the number reads in clusters
readNumbers( trimmedClusters)
The final step is folding, this step populates the
viennaStructures
, dgs
and
interactionTable
slots. This step can only be run if you
have the Vienna package installed and RNAFold in your PATH.
# Fold the RNA in part of whole
= foldComrades(trimmedClusters,
foldedCds rnaRefs = rnaRefs,
start = 1600,
end = 1869,
shape = 0,
ensembl = 20,
constraintNumber = 30,
evCutoff = 5)
# Check status of instance
foldedCds
Plots can be made for each sample using the plotMatrices function.
# Plot heatmaps for each sample
plotMatrices(cds = cds, # The comradesDataSet instance
type = "original", # The "analysis stage"
directory = 0, # The directory for output (0 for standard out)
a = 1, # Start coord for x-axis
b = rnaSize(cds), # End coord for x-axis
c = 1, # Start coord for y-axis
d = rnaSize(cds), # End coord for y-axis
h = 5) # The hight of the image (if saved)
See which contact maps are available by checking the status of the object
trimmedClusters#> comradesDataSet Object
#> RNAs Analysed - - ENSG000000XXXXX_NR003286-2_RN18S1_rRNA
#> Samples Analysed - - s1 c1
#> hybFiles - Raw data - original host noHost
#> matrixList - Contact Matricies - noHost original originalClusters superClusters trimmedClusters
#> clusterTableList - Cluster positions - original superClusters trimmedClusters
#> clusterGrangesList - Granges of clusters - original superClusters trimmedClusters
#> interactionTable - Predictied interaction - 0
#> viennaStructures - Predicted Structures - 0
# Plot heatmaps for all samples combined and all controls combined
plotMatricesAverage(cds = trimmedClusters, # The comradesDataSet instance
type = "trimmedClusters", # The "analysis stage"
directory = 0, # The directory for output (0 for standard out)
a = 1, # Start coord for x-axis
b = rnaSize(cds), # End coord for x-axis
c = 1, # Start coord for y-axis
d = rnaSize(cds), # End coord for y-axis
h = 5) # The hight of the image (if saved)
With large RNAS (?500bp), it can be useful to segment the RNA and fold the segments seaparately. DNA and RNA that form secondary and tertiary structures often have domains where there is more inter-domain interactions that inra-domain interactions. The TopDom package was designed to identify these domains for HI-C data. Using this package you can identify domains in the RNA structural data and can be used to inform the folding.
= data.frame()
domainDF for(j in c(20,30,40,50,60,70)){
#for(i in which(sampleTable(cds)$group == "s")){
= as.matrix(getData(x = cds,
timeMats data = "matrixList",
type = "noHost")[[1]])
= timeMats/ (sum(timeMats)/1000000)
timeMats = tempfile()
tmp write.table(timeMats, file = tmp,quote = F,row.names = F, col.names = F)
= readHiC(
tdData2 file = tmp,
chr = "rna18s",
binSize = 10,
debug = getOption("TopDom.debug", FALSE)
)
= TopDom(
tdData
tdData2 ,window.size = j,
outFile = NULL,
statFilter = TRUE,
debug = getOption("TopDom.debug", FALSE)
)
= tdData$domain
td $sample = sampleTable(cds)$sampleName[1]
td$window = j
td= rbind.data.frame(td, domainDF)
domainDF
}
ggplot(domainDF) +
geom_segment(aes(x = from.coord/10,
xend = to.coord/10, y = as.factor(sub("s","",sample)),
yend = (as.factor(sub("s","",sample)) ), colour = tag),
size = 20, alpha = 0.8) +
facet_grid(window~.)+
theme_bw()
A PCA of the structural ensembl can be made.
plotEnsemblePCA(foldedCds,
labels = T, # plot labels for structures
split = T) # split samples over different facets (T/f)
Compare two structures from the ensembl
plotComparisonArc(foldedCds = foldedCds,
s1 = "c1", # The sample of the 1st structure
s2 = "s1", # The sample of the 2nd structure
n1 = 13, # The number of the 1st structure
n2 = 16) # The number of the 2nd structure
plotStructure(foldedCds = foldedCds,
rnaRefs = rnaRefs,
s = "s1", # The sample of the structure
n = 1) # The number of the structure
Along with the RNA of interest the data also contains inter-RNA interactions with other RNAs from the transcriptome reference. After identifying abundant interactions using topInteractions you can find out where on each RNA these inetractions occur using getInteractions and getReverseInteractions.
getInteractions(cds,
"ENSG00000XXXXXX_NR003287-2_RN28S1_rRNA") %>%
mutate(sample =sub("\\d$","",sample) )%>%
group_by(rna,Position,sample)%>%
summarise(sum = sum(depth)) %>%
ggplot()+
geom_area(aes(x = Position,
y = sum,
fill = sample),
stat = "identity")+
facet_grid(sample~.) +
theme_bw()
#> `summarise()` has grouped output by 'rna', 'Position'. You can override using
#> the `.groups` argument.
getReverseInteractions(cds,
%>%
rna) mutate(sample =sub("\\d$","",sample) )%>%
group_by(rna,Position,sample)%>%
summarise(sum = sum(depth)) %>%
ggplot()+
geom_area(aes(x = Position,
y = sum,
fill = sample),
stat = "identity")+
facet_grid(sample~.)+
theme_bw()
#> `summarise()` has grouped output by 'rna', 'Position'. You can override using
#> the `.groups` argument.
The clusters can be compared to set of interactions to see which clusters share coordinates with a this set of interactions. The table should be formatted as a tabale fame of 2 columns (i and j) each colunn containing numerical values giving an interaction between i and j with which the clusters should be compared.
To compare to set of know interactions you need a contact matrix these interactions, for plotting it is sometimes useful to expand the interactions so they can be seen easily.
= 5
expansionSize = matrix(0, nrow = rnaSize(cds), ncol = rnaSize(cds))
knownMat for(i in 1:nrow(known18S)){
$V1[i]-expansionSize):(known18S$V1[i]+expansionSize),
knownMat[ (known18S$V2[i]-expansionSize):(known18S$V2[i]+expansionSize)] =
(known18S$V1[i]-expansionSize):(known18S$V1[i]+expansionSize),
knownMat[(known18S$V2[i]-expansionSize):(known18S$V2[i]+expansionSize)] +1
(known18S
}= knownMat + t(knownMat) knownMat
Using compareKnown you can check which clusters agree with the set of interactions. This functions adds analysis stages “known”, “novel” and “knownAndNovel” to the objects data attributes.
# use compare known to gett he known and not know clusters
= compareKnown(trimmedClusters, # The comradesDataSet instance
knowClusteredCds # A contact matrix of know interactions
knownMat, "trimmedClusters") # The analysis stage of clustering to compare
knowClusteredCds#> comradesDataSet Object
#> RNAs Analysed - - ENSG000000XXXXX_NR003286-2_RN18S1_rRNA
#> Samples Analysed - - s1 c1
#> hybFiles - Raw data - original host noHost
#> matrixList - Contact Matricies - noHost original originalClusters superClusters trimmedClusters KnownAndNovel novel known
#> clusterTableList - Cluster positions - original superClusters trimmedClusters novel known
#> clusterGrangesList - Granges of clusters - original superClusters trimmedClusters
#> interactionTable - Predictied interaction - 0
#> viennaStructures - Predicted Structures - 0
You can plot these using the plotMatrices function
# Plot heatmaps for all samples combined and all controls combined
plotMatricesAverage(cds = knowClusteredCds, # The comradesDataSet instance
type = "KnownAndNovel", # The "analysis stage"
directory = 0, # The directory for output (0 for standard out)
a = 1, # Start coord for x-axis
b = rnaSize(cds), # End coord for x-axis
c = 1, # Start coord for y-axis
d = rnaSize(cds), # End coord for y-axis
h = 5) # The hight of the image (if saved)
# Get the number of clusters for each analysis Stage
clusterNumbers(knowClusteredCds)
# Get the number of reads in each cluster for each analysis stage
readNumbers(knowClusteredCds)
To compare predicted structures with the know stucture use “compareKnownStructures”. This will give you the number of base pairs that agree between the ensembl of predicted structures and the structure imputted for comparison. This can be for better viewing.
head(compareKnownStructures(foldedCds,
# the comarison set known18S))
ggplot(compareKnownStructures(foldedCds, known18S)) +
geom_hline(yintercept = c(0.5,0.25,0.75,0,1),
colour = "grey",
alpha = 0.2)+
geom_vline(xintercept = c(0.5,0.25,0.75,0,1),
colour = "grey",
alpha = 0.2)+
geom_point(aes(x = sensitivity,
y = precision,
size = as.numeric(as.character(unlist(foldedCds@dgs))),
colour = str_sub(structureID,
start = 1 ,
end = 2))) +
xlim(0,1)+
ylim(0,1)+
theme_classic()
The package ships with a subsetted version of the unenriched dataset
which can be used to follow the vignette, the full dataset can be found
here: Un-enriched
rRNA dataset. There is also a function
makeExampleComradesDataSet()
from the comradesOO package
that will create a simple toy example comradesDataSet object, follow
this small section to get a feel for the package and it’s functionality.
The dataset has 1 sample consisting of a control and sample.
# make an example dataset
= makeExampleComradesDataSet()
cds cds
By plotting the reads on a 2D heatmap you can see structured areas of the transcript and long / short distance inter-RNA interactions. The “type” argument can be changed to plot any analysis stage in the matrixList slot which will become populated as the analysis proceeds.
# Have a look at the reads for each sample
plotMatricesAverage(cds = cds,
type = "original",
directory = 0,
a = 1,
b = rnaSize(cds),
c = 1,
d = rnaSize(cds),
h = 5)
Find the RNAs that interact with the RNA of interest.
topInteracters(cds,2)
Get some more information about the RNAs in the sample and thier abundance in each sample. With more RNAS and transcripts this becomes very useful as a count matrix.
featureInfo(cds)
Clustering and cluster trimming.
= clusterComrades(cds = cds,
clusteredCds cores = 1,
stepCount = 2,
clusterCutoff = 0)
= trimClusters(clusteredCds = clusteredCds,
trimmedClusters trimFactor = 1,
clusterCutoff = 0)
Plot them to check the clustering and trimming is working as expected.
plotMatricesAverage(cds = clusteredCds,
type = "originalClusters",
directory = 0,
a = 1,
b = rnaSize(cds),
c = 1,
d = rnaSize(cds),
h = 5)
plotMatricesAverage(cds = trimmedClusters,
type = "trimmedClusters",
directory = 0,
a = 1,
b = rnaSize(cds),
c = 1,
d = rnaSize(cds),
h = 5)
Folding:
= paste(c(rep('A',25),
fasta rep('T',25),
rep('A',10),
rep('T',23)),collapse = "")
= '>transcript1'
header
= tempfile()
fastaFile writeLines(paste(header,fasta,sep = "\n"),con = fastaFile)
= list()
rnaRefs rnas(cds)]] = read.fasta(fastaFile)
rnaRefs[[
rnaRefs
= foldComrades(trimmedClusters,
foldedCds rnaRefs = rnaRefs,
start = 1,
end = 83,
shape = 0,
ensembl = 5,
constraintNumber = 1,
evCutoff = 1)
If you have a set of nucloetide contacts, you can check to see if
they exist in the set of predicted structures after folding. Also see
compareKnown
which is a similar function but tests the set
of interactions against the clusters.
# make an example table of "know" interactions
= data.frame(V1 = c(6),
file V2 = c(80))
compareKnownStructures(foldedCds,file)
Plot a PCA of the predicted structures
plotEnsemblePCA(foldedCds, labels = T,split = T)
Compare two structures
plotComparisonArc(foldedCds = foldedCds,"s1","c1",2,3)
Plot one structure
plotStructure(foldedCds = foldedCds,"c1",1)