Skip to contents

Filter cells based on the number of genes, counts, and mitochondrial reads.

filter_cells.Rd

Filtering runs in two stages, and the log records how many cells each one removed.

Usage

filter_cells(
  x,
  num_mads = 3,
  min_genes = 500,
  max_genes = NULL,
  min_counts = NULL,
  max_counts = NULL,
  max_mt = 10,
  min_cells = 50,
  log_file = NULL
)

Arguments

x

A tibble with metadata (default method) or a Seurat object (Seurat method). Must have detected_genes/total_counts columns, as added by initialize_seurat_object().

num_mads

Median absolute deviations for the outlier stage. NULL skips the stage.

min_genes

Minimum number of genes per cell. NULL for no cutoff.

max_genes

Maximum number of genes per cell. NULL for no cutoff.

min_counts

Minimum number of counts per cell. NULL for no cutoff.

max_counts

Maximum number of counts per cell. NULL for no cutoff.

max_mt

Maximum percentage of mitochondrial reads per cell. NULL for no cutoff.

min_cells

Minimum number of cells that must survive filtering. Below this, filtering stop()s rather than handing back an object too small to analyze meaningfully. NULL skips the check.

log_file

Log file.

Value

Filtered data. The Seurat method also writes a pre-filtering QC violin plot (via plot_metrics_distribution()) and the unfiltered metadata (via save_metadata(), as metadata-unfiltered.csv.gz), plus the post-filtering QC plots (via plot_metrics_distribution() and plot_metrics_correlations()) to the working directory.

Details

  1. Outlier removal. Cells further than num_mads median absolute deviations from the median number of genes or counts are dropped. The median and MAD are computed on the log scale and the bounds back-transformed, which suits these right-skewed counts. num_mads = NULL skips it.

  2. Fixed cutoffs. min_genes / max_genes / min_counts / max_counts are applied to whatever survived stage 1, then max_mt. Each is skipped when NULL, so out of the box only min_genes (500) and max_mt (10) do anything.

Applying the fixed cutoffs second makes them a quality floor the data-driven bounds cannot undercut: a wide or contaminated distribution can put the MAD bound below 500 genes, and those cells are still removed. All cutoffs are inclusive.

Genes that are no longer detected in any of the retained cells are dropped, since filtering cells leaves them as all-zero rows.