Skip to contents

Write a QFeatures object to an .h5mu file.

writeLinkH5MU.Rd

Inverse of readLinkH5MU(). MuData::writeH5MU() covers the MultiAssayExperiment skeleton; the feature graph held in AssayLinks is written separately to the global .varp as a p x p sparse adjacency matrix over the global .var index.

Usage

writeLinkH5MU(
  object,
  path,
  feature_mapping_key = "feature_mapping",
  overwrite = FALSE
)

Arguments

object

A QFeatures object prepared with prepareQFeatures().

path

Path of the .h5mu file to create.

feature_mapping_key

Key to store the feature graph under in .varp.

overwrite

Whether to replace path if it already exists.

Value

path, invisibly.

Limitations

Writing to .h5mu does not preserve all R data types and structures:

  • Date and date-time classes (Date, POSIXct, and POSIXlt) are not supported natively and must be converted before writing.

  • The distinction between numeric NA and NaN is lost: both are read back as NaN by readLinkH5MU().

  • Complex data structures in rowData or metadata are not fully supported. Matrix-valued rowData columns are dropped with a warning. Unsupported metadata objects, such as scpModel objects (class ScpModel), must be simplified before writing.

  • Unused factor levels are not preserved.

  • Feature row names must be globally unique across all sets. Duplicate row names cause an error.

  • Integer and logical columns containing NA in global colData, set-specific colData, or rowData are converted to doubles with a warning. Their original types are lost; logical values become 1 and 0, while missing values remain missing.

Use object <- prepareQFeatures(object) before writing to address some of these limitations. prepareQFeatures() converts date and date-time columns in global colData and set-specific rowData and colData to formatted character strings, converts scpModel objects in set metadata to lists, and prefixes feature row names with set names when needed to make them globally unique, updating the set links accordingly. It does not restore the original date classes, missing-value distinction, unused factor levels, or integer/logical column types when the file is read back.

Examples

data("feat3", package = "QFeatures")
preparedFeat3 <- prepareQFeatures(feat3)
#> Warning: Prefixed feature row names with assay names because they were not globally unique across the QFeatures object.
filePath <- tempfile(fileext = ".h5mu")
writeLinkH5MU(preparedFeat3, filePath)
file.exists(filePath)
#> [1] TRUE
unlink(filePath)