In schifferl/tuberculosis: Tuberculosis Gene Expression Data for Machine Learning

tuberculosis

The r BiocStyle::Biocpkg("tuberculosis") R/Bioconductor package features tuberculosis gene expression data for machine learning. All human samples from GEO that did not come from cell lines, were not taken postmortem, and did not feature recombination have been included. The package has more than 10,000 samples from both microarray and sequencing studies that have been processed from raw data through a hyper-standardized, reproducible pipeline.

The Pipeline

To fully understand the provenance of data in the r BiocStyle::Biocpkg("tuberculosis") R/Bioconductor package, please see the tuberculosis.pipeline GitHub repository; however, all users beyond the extremely curious can ignore these details without consequence. Yet, a brief summary of data processing is appropriate here. Microarray data were processed from raw files (e.g. CEL files) and background corrected using the normal-exponential method and the saddle-point approximation to maximum likelihood as implemented in the r BiocStyle::Biocpkg("limma") R/Bioconductor package; no normalization of expression values was done; where platforms necessitated it, the RMA (robust multichip average) algorithm without background correction or normalization was used to generate an expression matrix. Sequencing data were processed from raw files (i.e. fastq files) using the nf-core/rnaseq pipeline inside a Singularity container; the GRCh38 genome build was used for alignment. Gene names for both microarray and sequencing data are HGNC-approved GRCh38 gene names from the genenames.org REST API.

Installation

To install r BiocStyle::Biocpkg("tuberculosis") from Bioconductor, use r BiocStyle::CRANpkg("BiocManager") as follows.

BiocManager::install("tuberculosis")

To install r BiocStyle::Biocpkg("tuberculosis") from GitHub, use r BiocStyle::CRANpkg("BiocManager") as follows.

BiocManager::install("schifferl/tuberculosis", dependencies = TRUE, build_vignettes = TRUE)

Most users should simply install r BiocStyle::Biocpkg("tuberculosis") from Bioconductor.

Load Package

To use the package without double colon syntax, it should be loaded as follows.

library(tuberculosis)

The package is lightweight, with few dependencies, and contains no data itself.

Finding Data

To find data, users will use the tuberculosis function with a regular expression pattern to list available resources. The resources are organized by GEO series accession numbers. If multiple platforms were used in a single study, the platform accession number follows the series accession number and is separated by a dash. The date before the series accession number denotes the date the resource was created.

tuberculosis("GSE103147")

The function will print the names of matching resources as a message and return them invisibly as a character vector. To see all available resources use "." for the pattern argument.

Getting Data

To get data, users will also use the tuberculosis function, but with an additional argument, dryrun = FALSE. This will either download resources from r BiocStyle::Biocpkg("ExperimentHub") or load them from the user's local cache. If a resource has multiple creation dates, the most recent is selected by default; add a date to override this behavior.

tuberculosis("GSE103147", dryrun = FALSE)

The function returns a list of SummarizedExperiment objects, each with a single assay, exprs, where the rows are features (genes) and the columns are observations (samples). If multiple resources are requested, multiple resources will be returned, each as a list element.

tuberculosis("GSE10799.", dryrun = FALSE)

The assay of each SummarizedExperiment object is named exprs rather than counts because it can come from either a microarray or a sequencing platform. If colnames begin with GSE, data comes from a microarray platform; if colnames begin with SRR, data comes from a sequencing platform.

No Metadata?

The SummarizedExperiment objects do not have sample metadata as colData, and this limits their use to unsupervised analyses for the time being. Sample metadata are currently undergoing manual curation, with the same level of diligence that was applied in data processing, and will be included in the package when they are ready.

Contributing

To contribute to the r BiocStyle::Biocpkg("tuberculosis") R/Bioconductor package, first read the contributing guidelines and then open an issue. Also note that in contributing you agree to abide by the code of conduct.

schifferl/tuberculosis documentation built on May 12, 2024, 1:57 p.m.