filter_low_coverage: Remove high-missingness features from an HDF5Matrix

In BigDataStatMeth: Scalable Statistical Computing with HDF5-Backed Matrices

Remove high-missingness features from an HDF5Matrix

Description

Removes columns (SNPs) or rows (samples) that carry too many missing values. Writes the result to a new dataset.

Missing values are the entries equal to 3, the missing-data code of the 0/1/2/3 genotype encoding shared by the omics functions; NA or NaN entries are not counted, so recode them to 3 beforehand. A column (or row) is removed when its proportion of 3s is greater than or equal to pcent.

Row and column names of x, when present, are carried over to the result: the names of the filtered axis are subset to the surviving elements.

When out_group/out_dataset are NULL (default) the result is written alongside the input dataset with the suffix "_filtered".

Usage

filter_low_coverage(x, ...)

## S3 method for class 'HDF5Matrix'
filter_low_coverage(
  x,
  out_group = NULL,
  out_dataset = NULL,
  pcent = 0.05,
  by_cols = TRUE,
  overwrite = FALSE,
  ...
)

Arguments

x	An HDF5Matrix containing SNP data with missing values coded as 3.
...	Ignored.
out_group	Output group. NULL (default) = same group as input.
out_dataset	Output dataset name. NULL (default) = input name + "_filtered".
pcent	Numeric in [0,1]. Missing-data threshold (default 0.05). Features whose proportion of 3s reaches this value are removed.
by_cols	Logical. Filter columns (TRUE, default) or rows.
overwrite	Logical. Overwrite existing output. Default FALSE.

Value

HDF5Matrix pointing to the filtered dataset, carrying the dimension names of the surviving elements.

Examples

set.seed(42)
fn <- tempfile(fileext = ".h5")
# 0/1/2 genotypes with about 10% missing values, coded as 3
snps <- matrix(sample(c(0, 1, 2, 3), 200, replace = TRUE,
                       prob = c(.3, .3, .3, .1)), 20, 10)
X   <- hdf5_create_matrix(fn, "geno/raw", data = snps)

# Filter with auto output path (adds "_filtered" suffix)
out <- filter_low_coverage(X, pcent = 0.2)

# Filter with explicit output
out2 <- filter_low_coverage(X, pcent = 0.2, out_group = "geno",
                             out_dataset = "filtered", overwrite = TRUE)
hdf5_close_all()
unlink(fn)

BigDataStatMeth documentation built on Sept. 15, 2026, 1:08 a.m.