tximport: Import transcript-level abundances and counts for transcript-...

In mikelove/tximport: Import and summarize transcript-level estimates for transcript- and gene-level analysis

Import transcript-level abundances and counts for transcript- and gene-level analysis packages

Description

tximport imports transcript-level estimates from various external software and optionally summarizes abundances, counts, and transcript lengths to the gene-level (default) or outputs transcript-level matrices (see txOut argument).

Usage

tximport(
  files,
  type = c("none", "salmon", "sailfish", "alevin", "piscem", "oarfish", "kallisto",
    "rsem", "stringtie"),
  txIn = TRUE,
  txOut = FALSE,
  countsFromAbundance = c("no", "scaledTPM", "lengthScaledTPM", "dtuScaledTPM"),
  tx2gene = NULL,
  varReduce = FALSE,
  dropInfReps = FALSE,
  infRepStat = NULL,
  ignoreTxVersion = FALSE,
  ignoreAfterBar = FALSE,
  geneIdCol,
  txIdCol,
  abundanceCol,
  countsCol,
  lengthCol,
  importer = NULL,
  existenceOptional = FALSE,
  sparse = FALSE,
  sparseThreshold = 1,
  readLength = 75,
  alevinArgs = NULL
)

Arguments

files	a character vector of filenames for the transcript-level abundances
type	character, the type of software used to generate the abundances. Options are "salmon", "sailfish", "alevin", "piscem", "oarfish", "kallisto", "rsem", "stringtie", or "none". This argument is used to autofill the arguments below (geneIdCol, etc.) "none" means that the user will specify these columns. Be aware that specifying type other than "none" will ignore the arguments below (geneIdCol, etc.)
txIn	logical, whether the incoming files are transcript level (default TRUE)
txOut	logical, whether the function should just output transcript-level (default FALSE)
countsFromAbundance	character, either "no" (default), "scaledTPM", "lengthScaledTPM", or "dtuScaledTPM". Whether to generate estimated counts using abundance estimates: scaled up to library size (scaledTPM), scaled using the average transcript length over samples and then the library size (lengthScaledTPM), or scaled using the median transcript length among isoforms of a gene, and then the library size (dtuScaledTPM). dtuScaledTPM is designed for DTU analysis in combination with txOut=TRUE, and it requires specifing a tx2gene data.frame. dtuScaledTPM works such that within a gene, values from all samples and all transcripts get scaled by the same fixed median transcript length. If using scaledTPM, lengthScaledTPM, or geneLengthScaledTPM, the counts are no longer correlated across samples with transcript length, and so the length offset matrix should not be used.
tx2gene	a two-column data.frame linking transcript id (column 1) to gene id (column 2). the column names are not relevant, but this column order must be used. this argument is required for gene-level summarization, and the tximport vignette describes how to construct this data.frame (see Details below). An automated solution to avoid having to create tx2gene if one has quantified with Salmon or alevin with human or mouse transcriptomes is to use the tximeta function from the tximeta Bioconductor package.
varReduce	whether to reduce per-sample inferential replicates information into a matrix of sample variances variance (default FALSE). alevin computes inferential variance by default for bootstrap inferential replicates, so this argument is ignored/not necessary
dropInfReps	whether to skip reading in inferential replicates (default FALSE). For alevin, tximport will still read in the inferential variance matrix if it exists
infRepStat	a function to re-compute counts and abundances from the inferential replicates, e.g. matrixStats::rowMedians to re-compute counts as the median of the inferential replicates. The order of operations is: first counts are re-computed, then abundances are re-computed. Following this, if countsFromAbundance is not "no", tximport will again re-compute counts from the re-computed abundances. infRepStat should operate on rows of a matrix. (default is NULL)
ignoreTxVersion	logical, whether to split the tx id on the '.' character to remove version information to facilitate matching with the tx id in tx2gene (default FALSE)
ignoreAfterBar	logical, whether to split the tx id on the '|' character to facilitate matching with the tx id in tx2gene (default FALSE). if txOut=TRUE it will strip the text after '|' on the rownames of the matrices
geneIdCol	name of column with gene id. if missing, the tx2gene argument can be used. Note that this argument and the other four "...Col" arguments below are ignored unless type="none"
txIdCol	name of column with tx id
abundanceCol	name of column with abundances (e.g. TPM or FPKM)
countsCol	name of column with estimated counts
lengthCol	name of column with feature length information
importer	a function used to read in the files
existenceOptional	logical, should tximport not check if files exist before attempting import (default FALSE, meaning files must exist according to file.exists)
sparse	logical, whether to try to import data sparsely (default is FALSE). Initial implementation for txOut=TRUE, countsFromAbundance="no" or "scaledTPM", no inferential replicates. Only counts matrix is returned (and abundance matrix if using "scaledTPM")
sparseThreshold	the minimum threshold for including a count as a non-zero count during sparse import (default is 1)
readLength	numeric, the read length used to calculate counts from StringTie's output of coverage. Default value (from StringTie) is 75. The formula used to calculate counts is: cov * transcript length / read length
alevinArgs	named list, with logical elements filterBarcodes, tierImport, forceSlow, dropMeanVar. See Details for definitions.

Details

Inferential replicates: tximport will also load in information about inferential replicates – a list of matrices of the Gibbs samples from the posterior, or bootstrap replicates, per sample – if these data are available in the expected locations relative to the files. The inferential replicates, stored in infReps in the output list, are on estimated counts, and therefore follow counts in the output list. By setting varReduce=TRUE, the inferential replicate matrices will be replaced by a single matrix with the sample variance per transcript/gene and per sample.

summarizeToGene: While tximport summarizes to the gene-level by default, the user can also perform the import and summarization steps manually, by specifing txOut=TRUE and then using the function summarizeToGene. Note however that this is equivalent to tximport with txOut=FALSE (the default).

Solutions on summarization: regarding "tximport failed at summarizing to the gene-level":

1. provide a tx2gene data.frame linking transcripts to genes (more below)

2. avoid gene-level summarization by specifying txOut=TRUE

See vignette('tximport') for example code for generating a tx2gene data.frame from a TxDb object. The tx2gene data.frame should exactly match and be derived from the same set of transcripts used for quantifying (the set of transcript used to create the transcriptome index).

Tximeta: One automated solution for Salmon/alevin/piscem/oarfish quantification data is to use the tximeta function in the tximeta Bioconductor package which builds upon and extends tximport; this solution should work out-of-the-box for human and mouse transcriptomes downloaded from GENCODE, Ensembl, or RefSeq. For other cases, the user should create the tx2gene manually as shown in the tximport vignette.

On tx2gene construction: Note that the keys and select functions used to create the tx2gene object are documented in the man page for AnnotationDb-class objects in the AnnotationDbi package (TxDb inherits from AnnotationDb). For further details on generating TxDb objects from various inputs see vignette('GenomicFeatures') from the GenomicFeatures package.

alevin: The alevinArgs argument includes some alevin-specific arguments. This optional argument is a list with any or all of the following named logical variables: filterBarcodes, tierImport, and forceSlow. The variables are described as follows (with default values in parens): filterBarcodes (FALSE) import only cell barcodes listed in whitelist.txt; tierImport (FALSE) import the tier information in addition to counts; forceSlow (FALSE) force the use of the slower import R code even if eds is installed; dropMeanVar (FALSE) don't import inferential mean and variance matrices even if they exist (also skips inferential replicates) For type="alevin" all arguments other than files, dropInfReps, and alevinArgs are ignored. Note that files should point to a single quants_mat.gz file, in the directory structure created by the alevin software (e.g. do not move the file or delete the other important files). Note that importing alevin quantifications will be much faster by first installing the eds package, which contains a C++ importer for alevin's EDS format. For alevin, tximport is importing the gene-by-cell matrix of counts, as txi$counts, and effective lengths are not estimated. txi$mean and txi$variance may also be imported if inferential replicates were used, as well as inferential replicates if these were output by alevin. Length correction should not be applied to datasets where there is not an expected correlation of counts and feature length.

Value

A simple list containing matrices: abundance, counts, length. Another list element 'countsFromAbundance' carries through the character argument used in the tximport call. The length matrix contains the average transcript length for each gene which can be used as an offset for gene-level analysis. If detected, and txOut=TRUE, inferential replicates for each sample will be imported and stored as a list of matrices, itself an element infReps in the returned list. An exception is alevin, in which the infReps are a list of bootstrap replicate matrices, where each matrix has genes as rows and cells as columns. If varReduce=TRUE the inferential replicates will be summarized according to the sample variance, and stored as a matrix variance. alevin already computes the variance of the bootstrap inferential replicates and so this is imported without needing to specify varReduce=TRUE.

References

Charlotte Soneson, Michael I. Love, Mark D. Robinson (2015) Differential analyses for RNA-seq: transcript-level estimates improve gene-level inferences. F1000Research. http://doi.org/10.12688/f1000research.7563

Examples

# load data for demonstrating tximport
# note that the vignette shows more examples
# including how to read in files quickly using the readr package

library(tximportData)
dir <- system.file("extdata", package="tximportData")
samples <- read.table(file.path(dir,"samples.txt"), header=TRUE)
files <- file.path(dir,"salmon", samples$run, "quant.sf.gz")
names(files) <- paste0("sample",1:6)

# tx2gene links transcript IDs to gene IDs for summarization
tx2gene <- read.csv(file.path(dir, "tx2gene.gencode.v27.csv"))

txi <- tximport(files, type="salmon", tx2gene=tx2gene)

mikelove/tximport documentation built on March 28, 2024, 5:05 p.m.