In Bioconductor/ShortRead: FASTQ input and manipulation

knitr::opts_chunk$set(collapse = TRUE)

library("ShortRead")

The r Biocpkg("ShortRead") package provides functionality for working with FASTQ files from high throughput sequence analysis. The package also contains functions for legacy (single-end, ungapped) aligned reads; for working with BAM files, please see the r Biocpkg("Rsamtools"), r Biocpkg("GenomicRanges"), r Biocpkg("GenomicAlignments") and related packages.

Sample data

Sample FASTQ data are derived from ArrayExpress record E-MTAB-1147. Paired-end FASTQ files were retrieved and then sampled to 20,000 records with

sampler <- FastqSampler('E-MTAB-1147/fastq/ERR127302_1.fastq.gz', 20000)
set.seed(123); ERR127302_1 <- yield(sampler)
sampler <- FastqSampler('E-MTAB-1147/fastq/ERR127302_2.fastq.gz', 20000)
set.seed(123); ERR127302_2 <- yield(sampler)

Functionality

Functionality is summarized in Table \@ref(tab:table).

Input FASTQ files are large so processing involves iteration in 'chunks' using FastqStreamer

strm <- FastqStreamer("a.fastq.gz")
repeat {
    fq <- yield(strm)
    if (length(fq) == 0)
        break
    ## process chunk
}

or drawing a random sample from the file

sampler <- FastqSampler("a.fastq.gz")
fq <- yield(sampler)

The default size for both streams and samples is 1M records; this volume of data fits easily into memory. Use countFastq to get a quick and memory-efficient count of the number of records and nucleotides in a file

: (#tab:table) Key functions for working with FASTQ files

| Input | | ---------------------|------------------------------------------- | | FastqStreamer | Iterate through FASTQ files in chunks | | FastqSampler | Draw random samples from FASTQ files | | readFastq | Read an entire FASTQ file into memory | | writeFastq | Write FASTQ objects to a connection (file) | | countFastq | Quickly count FASTQ records in files |

| Sequence and quality summary | | ---------------------|------------------------------------------- | | alphabetFrequency | Nucleotide or quality score use per read | | alphabetByCycle | Nucleotide or quality score use by cycle | | alphabetScore | Whole-read quality summary | | encoding | Character / ‘phred’ score mapping |

| Quality assessment | | ---------------------|------------------------------------------- | | qa | Visit FASTQ files to collect QA statistics | | report | Generate a quality assessment report |

| Filtering and trimming | | ---------------------|------------------------------------------- | | srFilter | Pre-defined and bespoke filters | | trimTails, etc. | Trim low-quality nucleotides | | narrow | Remove leading / trailing nucleotides | | tables | Summarize read occurrence | | srduplicated, etc. | Identify duplicate reads | | filterFastq | Filter reads from one file to another |

fl <- system.file(package="ShortRead", "extdata", "E-MTAB-1147",
                  "ERR127302_1_subset.fastq.gz")
countFastq(fl)

Small FASTQ files can be read into memory in their entirety using readFastq; we do this for our sample data

fq <- readFastq(fl)

The result of data input is an instance of class ShortReadQ (Table \@ref(tab:table2)). This class contains reads, their quality scores, and optionally the id of the read.

: (#tab:table2) Primary data types in the r Biocpkg("ShortRead") package.

| | | -------------------- | ------------------------------------------------- | | DNAStringSet | (r Biocpkg("Biostrings")) Short read sequences | | FastqQuality, etc. | Quality encodings | | ShortReadQ | Reads, quality scores, and ids |

fq
fq[1:5]
head(sread(fq), 3)
head(quality(fq), 3)

The reads are represented as DNAStringSet instances, and can be manipulated with the rich tools defined in the r Biocpkg("Biostrings") package. The quality scores are represented by a class that represents the quality encoding inferred from the file; the encoding in use can be discovered with

encoding(quality(fq))

The primary source of documentation for these classes is ?ShortReadQ and ?QualityScore.

Common workflows

Quality assessment

FASTQ files are often used for basic quality assessment, often to augment the purely technical QA that might be provided by the sequencing center with QA relevant to overall experimental design. A QA report is generated by creating a vector of paths to FASTQ files

fls <- dir("/path/to", "*fastq$", full=TRUE)

collecting statistics over the files

qaSummary <- qa(fls, type="fastq")

and creating and viewing a report

browseURL(report(qaSummary))

By default, the report is based on a sample of 1M reads.

These QA facilities are easily augmented by writing custom functions for reads sampled from files, or by exploiting the elements of the object returned from qa(), e.g., for an analysis of ArrayExpress experiment E-MTAB-1147:

load("qa_E-MTAB-1147.Rda")

qaSummary

For instance, the count of reads in each lane is summarized in the readCounts element, and can be displayed as

head(qaSummary[["readCounts"]])
head(qaSummary[["baseCalls"]])

The readCounts element contains a data frame with 1 row and 3 columns (these dimensions are indicated in the parenthetical annotation of readCounts in the output of qaSummary). The rows represent different lanes. The columns indicated the number of reads, the number of reads surviving the Solexa filtering criteria, and the number of reads aligned to the reference genome for the lane. The baseCalls element summarizes the base calls in the unfiltered reads.

The functions that produce the report tables and graphics are internal to the package. They can be accessed by calling ShortRead:::functionName where functionName is one of the functions listed below, organized by report section.

• Run Summary: .ppnCount, .df2a, .laneLbl, .plotReadQuality
• Read Distribution: .plotReadOccurrences, .freqSequences
• Cycle Specific: .plotCycleBaseCall, .plotCycleQuality
• Tile Performance: .atQuantile, .colorkeyNames, .plotTileLocalCoords, .tileGeometry, .plotTileCounts, .plotTileQualityScore
• Alignment: .plotAlignQuality
• Multiple Alignment: .plotMultipleAlignmentCount
• Depth of Coverage: .plotDepthOfCoverage
• Adapter Contamination: .ppnCount

Filtering and trimming

It is straightforward to create filters to eliminate reads or to trim reads based on diverse characteristics. The basic structure is to open a FASTQ file, iterate through chunks of the file, performing filtering or trimming steps, and appending the filtered data to a new file.

myFilterAndTrim <-
    function(fl, destination=sprintf("%s_subset", fl))
{
    ## open input stream
    stream <- open(FastqStreamer(fl))
    on.exit(close(stream))
    repeat {
        ## input chunk
        fq <- yield(stream)
        if (length(fq) == 0)
            break
        ## trim and filter, e.g., reads cannot contain 'N'...
        fq <- fq[nFilter()(fq)]  # see ?srFilter for pre-defined filters
        ## trim as soon as 2 of 5 nucleotides has quality encoding less
        ## than "4" (phred score 20)
        fq <- trimTailw(fq, 2, "4", 2)
        ## drop reads that are less than 36nt
        fq <- fq[width(fq) >= 36]
        ## append to destination
        writeFastq(fq, destination, "a")
    }
}

This is memory efficient and flexible. Care must be taken to coordinate pairs of FASTQ files representing paired-end reads, to preserve order.

Using ShortRead for data exploration

Data I/O

r Biocpkg("ShortRead") provides a variety of methods to read data into R, in addition to readAligned.

readXStringColumns

readXStringColumns reads a column of DNA or other sequence-like data. For instance, the Solexa files s_N_export.txt contain lines with the following information:

## location of file
exptPath <- system.file("extdata", package="ShortRead")
sp <- SolexaPath(exptPath)
pattern <- "s_2_export.txt"
fl <- file.path(analysisPath(sp), pattern)
strsplit(readLines(fl, n=1), "\t")
length(readLines(fl))

Column 9 is the read, and column 10 the ASCII-encoded Solexa Fastq quality score; there are 1000 lines (i.e., 1000 reads) in this sample file.

Suppose the task is to read column 9 as a DNAStringSet and column 10 as a BStringSet. DNAStringSet is a class that contains IUPAC-encoded DNA strings (IUPAC code allows for nucleotide ambiguity); BStringSet is a class that contains any character with ASCII code 0 through 255. Both of these classes are defined in the r Biocpkg("Biostrings") package. readXStringColumns allows us to read in columns of text as these classes.

Important arguments for readXStringColumns are the dirPath in which to look for files, the pattern of files to parse, and the colClasses of the columns to be parsed. The dirPath and pattern arguments are like list.files. colClasses is like the corresponding argument to read.table: it is a list specifying the class of each column to be read, or NULL if the column is to be ignored. In our case, there are 21 columns, and we would like to read in columns 9 and 10. Hence

colClasses <- rep(list(NULL), 21)
colClasses[9:10] <- c("DNAString", "BString")
names(colClasses)[9:10] <- c("read", "quality")

We use the class of the type of sequence (e.g., DNAString or BString), rather than the class of the set that we will create ( e.g., DNAStringSet or BStringSet). Applying names to colClasses is not required, but makes subsequent manipulation easier. We are now ready to read our file

cols <- readXStringColumns(analysisPath(sp), pattern, colClasses)
cols

The file has been parsed, and appropriate data objects were created.

A feature of readXStringColumns and other input functions in the r Biocpkg("ShortRead") package is that all files matching pattern in the specified dirPath will be read into a single object. This provides a convenient way to, for instance, parse all tiles in a Solexa lane into a single DNAStringSet object.

There are several advantages to reading columns as XStringSet objects. These are more compact than the corresponding character representation:

object.size(cols$read)
object.size(as.character(cols$read))

They are also created much more quickly. And the DNAStringSet and related classes are used extensively in r Biocpkg("ShortRead"), r Biocpkg("Biostrings"), r Biocpkg("BSgenome") and other packages relevant to short-read technology.

Sorting

Short reads can be sorted using srsort, or the permutation required to bring the short read into lexicographic order can be determined using srorder. These functions are different from sort and order because the result is independent of the locale, and they operate quickly on DNAStringSet and BStringSet objects.

The function srduplicated identifies duplicate reads. This function returns a logical vector, similar to duplicated. The negation of the result from srduplicated is useful to create a collection of unique reads. An experimental scenario where this might be useful is when the sample preparation involved PCR. In this case, replicate reads may be due to artifacts of sample preparation, rather than differential representation of sequence in the sample prior to PCR.

Summarizing read occurrence

The tables function summarizes read occurrences, for instance,

tbls <- tables(fq)
names(tbls)
tbls$top[1:5]
head(tbls$distribution)

The top component returned by tables is a list tallying the most commonly occurring sequences in the short reads. Knowledgeable readers will recognize the top-occurring read as a close match to one of the manufacturer adapters.

The distribution component returned by tables is a data frame that summarizes how many reads (e.g., r tbls[["distribution"]][1,"nReads"]) are represented exactly r tbls[["distribution"]][1,"nOccurrences"] times.

Finding near matches to short sequences

Facilities exist for finding reads that are near matches to specific sequences, e.g., manufacturer adapter or primer sequences. srdistance reports the edit distance between each read and a reference sequence. srdistance is implemented to work efficiently for reference sequences whose length is of the same order as the reads themselves (10's to 100's of bases). To find reads close to the most common read in the example above, one might say

dist <- srdistance(sread(fq), names(tbls$top)[1])[[1]]
table(dist)[1:10]

'Near' matches can be filtered, e.g.,

fqSubset <- fq[dist>4]

A different strategy can be used to tally or eliminate reads that consist predominantly of a single nucleotide. alphabetFrequency calculates the frequency of each nucleotide (in DNA strings) or letter (for other string sets) in each read. Thus one could identify and eliminate reads with more than 30 adenine nucleotides with

countA <- alphabetFrequency(sread(fq))[,"A"]
fqNoPolyA <- fq[countA < 30]

alphabetFrequency, which simply counts nucleotides, is much faster than srdistance, which performs full pairwise alignment of each read to the subject.

Users wanting to use R for whole-genome alignments or more flexible pairwise alignment are encouraged to investigate the r Biocpkg("Biostrings") and r Biocpkg("pwalign") packages, especially the PDict class and matchPDict and pairwiseAlignment functions.

Legacy support for early file formats

The r Biocpkg("ShortRead") package contains functions and classes to support early file formats and ungapped alignments. Help pages are flagged as 'legacy'; versions of r Biocpkg("ShortRead") prior to 1.21 (Bioconductor version 2.13) contains a vignette illustrating common workflows with these file formats.

Session Info{-}

sessionInfo()

Bioconductor/ShortRead documentation built on May 5, 2024, 4:14 a.m.