Skip to contents

GC-correct a NIPTeR sample or control group

nipter_gc_correct.Rd

Adjusts bin counts for GC-content bias using either LOESS regression or bin-weight normalisation.

Usage

nipter_gc_correct(
  object,
  fasta = NULL,
  method = c("loess", "bin"),
  span = 0.75,
  include_sex = FALSE,
  binsize = 50000L,
  gc_table = NULL,
  con = NULL
)

Arguments

object

A NIPTeRSample or NIPTeRControlGroup.

fasta

Path to an indexed reference FASTA file (.fa/.fasta with .fai index). Ignored when gc_table is supplied.

method

GC correction method: "loess" (default) or "bin" (bin-weight).

span

LOESS smoothing parameter (only used when method = "loess"). Default 0.75.

include_sex

Logical; correct sex chromosomes (X, Y) as well? Default FALSE.

binsize

Bin size used when binning the sample (default 50000L). Must match the binsize of the sample. Ignored when gc_table is a list (bin size is already encoded in the vector lengths).

gc_table

Pre-computed GC table. Either a path to a TSV.bgz file (from nipter_gc_precompute) or the in-memory named list returned by a previous .get_gc_table() call. When NULL (default), GC content is computed from fasta.

con

Optional open DBI connection with duckhts loaded. If NULL, a temporary connection is created.

Value

A corrected copy of object with the same class. Correction status is updated from "Uncorrected" to "GC corrected".

Details

GC content can be supplied in three ways via the gc_table parameter:

Pre-computed file

Path to a TSV.bgz produced by nipter_gc_precompute. Fastest for large cohorts — compute once, reuse for every sample.

In-memory list

Named list of numeric vectors (one per chromosome) as returned by .get_gc_table(). Useful when chaining corrections within a session.

FASTA path via fasta

Compute GC on-the-fly for every call. Convenient for single-sample use; slow for many samples.

LOESS method (default): Fits a LOESS curve of read counts vs GC percentage across all autosomal bins with known GC and non-zero reads. Each bin is then scaled by median(counts) / fitted(loess), so that all bins are normalised to the genome-wide median. This is the NIPTeR default method.

Bin-weight method: Groups bins by GC percentage (0.1\ computes the mean read count per GC bucket, then scales each bin by global_mean / bucket_mean. Faster than LOESS but less smooth.

Sex chromosome correction (when include_sex = TRUE) uses a nearest-neighbour lookup against the autosomal LOESS curve (LOESS method) or the same GC bucket weights (bin-weight method).

See also

nipter_gc_precompute(), nipter_bin_bam(), nipter_chi_correct(), nipter_as_control_group()

Examples

if (FALSE) { # \dontrun{
# One-shot: compute GC and correct in one call
corrected <- nipter_gc_correct(sample, fasta = "hg38.fa")

# Recommended for cohorts: precompute once, reuse
nipter_gc_precompute("hg38.fa", binsize = 50000L, out = "hg38_gc_50k.tsv.bgz")
cg <- nipter_gc_correct(cg, gc_table = "hg38_gc_50k.tsv.bgz")
test_sample <- nipter_gc_correct(test_sample, gc_table = "hg38_gc_50k.tsv.bgz")
} # }