Analyzing 3’-seq/Term-seq Data with PIPETS
Introduction
Bacteria rely on rapid alteration of transcription termination to regulate their response to envrionmental perturbations. Premature termination of transcripts as well as attenuation of transcript length (ex: only expressing some genes in an operon) are both prevalent signals and studying them has potent implications for better understanding how antimicrobials affect bacteria. High throughput sequencing for studying termination signal (known as 3’-seq or term-seq) is currently available and more data sets are being created rapidly. However, unlike other high throughput sequencing methods, there is not standardized, easily accessible, statistically robust analysis tool. Here we present PIPETS (Poisson Identification of PEaks from Term-Seq data) which analyzes 3’-seq data using a Poisson Distribution Test to identify termination signal that is statistically higher than the surrounding noise. These termination peaks can be used to identify shifts in bacterial response and give a better understanding of a very important set of targets for ongoing antimicrobial development.
Quick Start
Installation
To install PIPETS, use Bioconductor’s BiocManager package.
Basic PIPETS Run
PIPETS accepts two types of data as inputs: Bed Files or GRanges
objects. Both sets of data undergo the same analysis method. Running the
method differs slightly if the user is using a GRanges input. Here we
show the basic walkthrough of running PIPETS on an input Bed file. The
five parameters that need to be specified to run PIPETS are
inputData, the readScoreMinimum of “good
quality” reads,
the prefix for the output file name OutputFileID, the file
path to the output director OutputFileDir ,and the type of
input data inputDataFormat.
Bed File Input
## Warning: multiple methods tables found for 'intersect'## Warning: multiple methods tables found for 'union'## Warning: multiple methods tables found for 'intersect'## Warning: multiple methods tables found for 'setdiff'GRanges Object Input
library(PIPETS)
library(BiocGenerics)
library(GenomicRanges)
#GRanges object input
#When run on GRanges objects, PIPETS also outputs the strand split GRanges objects to a list for the user
#The first item in the list are the + strand reads, and the second are the - strand reads
GRanges_Object <- read.delim(file = "PIPETS_TestData.bed",
header = FALSE, stringsAsFactors = FALSE)
GRanges_Object <- GRanges(seqnames = GRanges_Object$V1,
ranges = IRanges(start = GRanges_Object$V2,end = GRanges_Object$V3
),score = GRanges_Object$V5 ,strand = GRanges_Object$V6)
ResultsList <- PIPETS_FullRun(inputData = GRanges_Object, readScoreMinimum = 30,
OutputFileDir = tempdir(),
OutputFileID = "ExampleResultsRun", inputDataFormat = "GRanges")
head(ResultsList)## [[1]]
## GRanges object with 908 ranges and 2 metadata columns:
## seqnames ranges strand | score coverage
## <Rle> <IRanges> <Rle> | <integer> <integer>
## 1 NC_003028 29924-29982 + | 42 1
## 2 NC_003028 29933-29991 + | 42 2
## 3 NC_003028 30309-30367 + | 38 1
## 4 NC_003028 30310-30368 + | 39 9
## 5 NC_003028 30311-30369 + | 39 21
## ... ... ... ... . ... ...
## 905 NC_003028 34657-34715 + | 42 1
## 906 NC_003028 34660-34718 + | 42 1
## 907 NC_003028 34663-34721 + | 42 2
## 908 NC_003028 34664-34722 + | 42 1
## 909 NC_003028 34665-34723 + | 42 1
## -------
## seqinfo: 1 sequence from an unspecified genome; no seqlengths
##
## [[2]]
## GRanges object with 547 ranges and 2 metadata columns:
## seqnames ranges strand | score coverage
## <Rle> <IRanges> <Rle> | <integer> <integer>
## 1 NC_003028 641-699 - | 42 1
## 2 NC_003028 642-700 - | 42 1
## 3 NC_003028 643-701 - | 42 1
## 4 NC_003028 648-706 - | 42 2
## 5 NC_003028 660-718 - | 42 7
## ... ... ... ... . ... ...
## 544 NC_003028 34661-34719 - | 42 2
## 545 NC_003028 34662-34720 - | 42 31
## 546 NC_003028 34663-34721 - | 42 133
## 547 NC_003028 34664-34722 - | 42 97
## 548 NC_003028 34665-34723 - | 42 6
## -------
## seqinfo: 1 sequence from an unspecified genome; no seqlengthsBed File Input and Strand Specific Analysis
In the event that there is a large disparity in read depth or the presence of outliers between the top and complement data of a given data set, PIPETS can use different values for threshAdjust and highOutlierTrim to account for each strand uniquely. In order to perform this analysis, threshAdjust and highOutlierTrim should each be set to NA and the strand specific values of each of those parameters should then be set to the desired values.
library(PIPETS)
#Bed File Input
PIPETS_FullRun(inputData = "PIPETS_TestData.bed"
,readScoreMinimum = 30, OutputFileID = "ExampleResultsRun",
OutputFileDir = tempdir(), threshAdjust = NA,
threshAdjust_TopStrand = 0.75, threshAdjust_CompStrand = 0.55,
highOutlierTrim = NA, highOutlierTrim_TopStrand = 0.01,
highOutlierTrim_CompStrand = 0.05,
inputDataFormat = "bedFile")PIPETS Output
Per input Bed file, PIPETS will create 4 output files in the R directory. All output files will begin with the user defined file name.
First, PIPETS creates strand-split Bed files which it uses during analysis. PIPETS takes the input Bed file, separates the Plus-Strand (+) and the Complement-Strand (-) reads and counts how many reads are assigned to each genomic coordinate.
For the results, PIPETS outputs a list of statistically significant termination peaks for each strand. The output files contain information about the entire termination peak: the coordinate with the highest termination read coverage, the termination read coverage at that position, as well as the coordinates spanning the full length of the termination peak. While the position of highest termination read coverage is the most important result per peak, we also wanted to recognize the shape of imperfect bacterial termination by including the full range of each peak.
Detailed Walkthrough
Term-seq/3’seq have allowed for the study of bacterial termination in a high throughput manner. While there is a growing amount of studies using this data to answer biological questions, there are still few robust analysis methods that are widely available for use. Also, these methods do not analyze all termination reads, focusing only on termination signal found inside of the 3’-UTR of genes.
PIPETS is the first term-seq/3’-seq analysis method that uses statistically robust analysis methods to differentiate significant termination signal from surrounding noise and provide termination results for all genomic contexts, not just regions near genes. PIPETS is also the first term-seq/3’-seq analysis method that is easily available on Bioconductor.
Input Data Specifications
PIPETS takes in aligned term-seq/3’-seq reads in Bed format. Traditional Bedformat files always has the same first 3 columns: chrom, chromStart, and chromStop. PIPETS requires that these three columns are in this order to function. PIPETS also requires the score column and the strand column. PIPETS automtically detects which column contains the strand information, so the user does not need to specify its location.
PIPETS intentionally does not use gene information during its analysis. This is to ensure that there is no gene-centric bias, and that all termination signal is treated equally during significance testing. PIPETS’ output files contain the genomic coordinates for all results, making downstream gene association easy.
Method Steps
PIPETS performs data analysis in three steps which takes raw Bed file term-seq/3’-seq reads and outputs a list of significant termination peaks for each strand. This section has a more detailed walkthrough of each step.
Step One: Poisson Test to Identify Significant Positions
The first step of PIPETS is to identify all genomic positions with termination read coverage that is statistically significantly higher than that of the surrounding area. We employ a Poisson Distribution Test in a sliding window framework to accomplish this. Using a Poisson Distribution Test allows for the assignment of statistical significance to our results. It also provides a way to identify biologically significant signal that does not rely solely on global read coverage cutoffs. While PIPETS does use a global cutoff, it is designed to capture very low read coverage values which are very likely to be biological or technical noise. Using a sliding window framework allows for a more sensitive identification system of significant read coverage levels. The sliding window moves through the input data and tests small regions of the data, ensuring that outlier read coverage values do not skew PIPETS ability to identify genomic positions with read coverage that is significant in relation to its surrounding region.
By default, PIPETS begins by creating the first sliding window at the 26th position in the input data that spans 25bp upstream and downstream. It then performs a Poisson Distribution Test on each position in the window, comparing its read coverage to the average read coverage of all positions in the window. If its coverage is significantly higher than expected from the average of the window, then it is identified as significant.
After all positions in the window have been tested, the window then moves downstream by 25 bp (by default) and tests again. This movement intentionally tests half of the positions a second time, increasing the potential for positions with read coverage that is overshadowed by nearby outliers to be tested again, increasing the sensitivity of PIPETS.
After the final sliding window has been tested, PIPETS compiles all genomic positions that were identified at least once as being significant. We then employ Benjamini-Hochberg multiple testing correction to reduce the incidence of Type I error. The resulting list of genomic positions with significant read coverage is then passed to downstream steps. Many of these positions will be consecutive with at least one other position, which is expected as bacterial termination is not a perfectly succinct process, and there is often signal surrounding true termination positions. The next step of PIPETS processes these positions and addresses the data architecture of termination “peaks”.
Step Two: Condensing of Proximal Significant Positions
Step two of PIPETS begins with the list of all genomic positions with termination read coverage that is significantly higher than their surrounding area. True termination signal in bacteria resembles peak like data structures, with a position with the most coverage (likely the true terminator) and positions on either side with termination read coverage that is higher than the surrounding positions, but less than that of the highest point. All of the positions inside of a peak likely have read coverages that are higher than the surrounding area, and will be marked as significant. But they are not all distinct termination signals, rather they are all part of the same termination peak. Step two of PIPETS moves through the results set from step one and combines significant positions that are within 2bp of each other into single termination peaks. The position with the highest read coverage is identified, and the 3’-most and 5’-most coordinates are saved. This both reduces the total number of results that need to be analyzed without discarding relevant data about the results.
Step Three: Condensing of Proximal Significant Peaks
Step three of PIPETS is similar to step two, but instead of condensing consecutive significant positions, it condenses proximal significant peaks. This is done for similar reasons as step two: we do not expect very proximal termination peaks to be two distinct signals, rather they are likely the result of loss of consistent signal due to technical problems. The default distance for this step is 20bp, and during parameter testing there were no identified cases of this distance causing PIPETS to combine two independent signals. After this step has completed running, PIPETS will output the strand specific results files.
Other Parameters
While PIPETS only requires 5 parameters from the user to run, there are other parameters that can be tuned to alter PIPETS’ performance. These parameters have set defaults intended to provide a moderately strict analysis.
slidingWindowSize
slidingWindowSize (numerical value) defines the total size of the sliding window that is used in the first step of PIPETS. The default is set to 25, which will create a sliding window of size 51 (the first sliding window is created on the 26th genomic position by default, and extends 25 bp upstream and downstream of that postition).
slidingWindowMovementDistance
slidingWindowMovementDistance (numerical value) defines the distance that the sliding window moves during the first step of PIPETS. By default it is set to 25. This ensures that almost every position in the input data is tested twice, increasing the likelihood that PIPETS identifies biologically significant termination signal that might otherwise be dwarfed by proximal positions with very high termination read coverage. In testing, adjusting the slidingWindowMovementDistance to be much smaller than the slidingWindowSize (which in turn would test most positions in the data more than two times) had diminishing returns on increasing the sensitivity, and caused higher run times.
adjacentPeakDistance
adjacentPeakDistance (numerical value) is used in the second step of PIPETS to combine signal from adjacent signficant termination positions. By default this is set to 2, meaning that if two significant positions are within 2 bp of each other, they are added to the same termination peak. In testing, raising this value had very little impact, but by keeping above 1, we are ensuring that loss of signal due to technical problems does not create situations in which very proximal signal is confused for separate termination results.
peakCondensingDistance
peakCondensingDistance (numerical value) is used in the third step of PIPETS to combine signal from proximal termination peaks. By default this is set to 20. It is used to test if peaks are close enough together to be from the same termination signal, or if they are sufficiently far away to be distinct termination sites.
threshAdjust
threshAdjust (numerical value greater than 0 and less than 1 as decimal) is used to establish a global cutoff for minimum read coverage for the input data. By default it is set to 0.75. This parameter describes the percentage of the total read coverage that should be used to inform an “average” read coverage cutoff. In other words, take the top 75% of the read coverage positions and assume that those positions encompass the biologically relevant read coverage levels (and are likely too high to be the result of biological or technical noise) that can be used to define an expected read coverage level for biologically significant signal.
user_pValue
user_pValue (numerical value less than 1 as decimal) is used in the first step of PIPETS and is the user defined p-value cutoff for the initial Poisson Distribution Test and the subsequent Benjamini-Hochberg multiple testing correction. By default it is set to 0.0005.
highOutlierTrim
highOutlierTrim (numerical value greater than 0 and less than 1 as decimal) is used to modulate the global read coverage minumum cutoff. By default it is set to 0.01. This parameter is used to reduce the influence of very high read coverage positions in the creation of the global cutoff. The parameter describes what total percentage of the highest read coverage positions should be excluded in the cutoff creation process. It removes the top 1% of positions (if there are 500 positions that contain the top 75% of all reads then this parameter states to remove the top 5 of those top positions).
inputDataFormat
The default value is bedFile which assumes that the
input will be a string file path to the input bed file for the run.
Alternatively, if the user has created a GRanges object, the user can
specify GRanges and enter the R GRanges object into the run
method.
Session Info
## R version 4.4.2 (2024-10-31)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 24.04.1 LTS
##
## Matrix products: default
## BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
## LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so; LAPACK version 3.12.0
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_US.UTF-8 LC_COLLATE=C
## [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## time zone: Etc/UTC
## tzcode source: system (glibc)
##
## attached base packages:
## [1] stats4 stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] GenomicRanges_1.59.0 GenomeInfoDb_1.43.0 IRanges_2.41.1
## [4] S4Vectors_0.45.2 BiocGenerics_0.53.3 generics_0.1.3
## [7] PIPETS_1.3.0 BiocStyle_2.35.0
##
## loaded via a namespace (and not attached):
## [1] jsonlite_1.8.9 dplyr_1.1.4 compiler_4.4.2
## [4] BiocManager_1.30.25 tidyselect_1.2.1 jquerylib_0.1.4
## [7] yaml_2.3.10 fastmap_1.2.0 R6_2.5.1
## [10] XVector_0.47.0 knitr_1.49 tibble_3.2.1
## [13] maketools_1.3.1 GenomeInfoDbData_1.2.13 bslib_0.8.0
## [16] pillar_1.9.0 rlang_1.1.4 utf8_1.2.4
## [19] cachem_1.1.0 xfun_0.49 sass_0.4.9
## [22] sys_3.4.3 cli_3.6.3 magrittr_2.0.3
## [25] zlibbioc_1.52.0 digest_0.6.37 lifecycle_1.0.4
## [28] vctrs_0.6.5 evaluate_1.0.1 glue_1.8.0
## [31] buildtools_1.0.0 fansi_1.0.6 rmarkdown_2.29
## [34] httr_1.4.7 pkgconfig_2.0.3 tools_4.4.2
## [37] htmltools_0.5.8.1 UCSC.utils_1.3.0