Struct QcConfig 

pub struct QcConfig {

    pub n_mads: f32,
    pub min_cell_nnz: usize,
    pub min_counts_per_cell: f32,
    pub mito_pattern: Option<String>,
    pub mito_max_frac: Option<f32>,
    pub ribo_pattern: Option<String>,
    pub ribo_max_frac: Option<f32>,
    pub mad_on_n_genes: bool,
    pub mad_on_counts: bool,
    pub mad_on_mito: bool,
    pub feature_min_cells: usize,
    pub drop_outliers: bool,
    pub auto_cell_cutoff: bool,
    pub qc_histogram: bool,
}

Configuration for cell QC. Plain struct (no clap) so non-clap crates can construct it directly; the clap surface is QcArgs.

Fields

n_mads: f32

MAD multiplier for the robust band (larger = more permissive).

min_cell_nnz: usize

Near-empty floor: cells with fewer than this many detected features (nnz across all rows) are masked at output, not dropped from training. 0 disables the floor.

min_counts_per_cell: f32

Hard floor on total counts per cell, in addition to MAD. 0 disables.

mito_pattern: Option<String>

Regex over row (feature) names selecting mitochondrial genes; enables the per-cell mito-fraction metric. None disables it.

mito_max_frac: Option<f32>

Optional hard max mitochondrial fraction (0..1).

ribo_pattern: Option<String>

Regex selecting ribosomal genes (enables ribo-fraction metric).

ribo_max_frac: Option<f32>

Optional hard max ribosomal fraction (0..1).

mad_on_n_genes: bool

Per-cell metrics that drive MAD outlier flagging.

mad_on_counts: bool

mad_on_mito: bool

feature_min_cells: usize

Feature-axis QC (off by default): drop genes expressed in fewer than this many cells. 0 disables.

drop_outliers: bool

Master switch for the MAD train-drop tier. When false, only the near-empty floor (output mask) is computed — used by inference (predict/impute) so query cells are never silently dropped.

auto_cell_cutoff: bool

Automatic cell calling: pick the per-cell nnz cutoff at the trough between the ambient and the cell peak (crate::qc::suggest_nnz_cutoff) and train-drop every cell below it. So called-out ambient is removed up front (via the caller’s mask_columns), not just output-masked — it never shapes the model. The cutoff is authoritative; min_cell_nnz does NOT floor it (it only drives the separate near-empty output mask). No-op when there is no trough (unimodal) or when drop_outliers is false (inference). Unlike the loader’s per-file empty-barcode gate, this runs on the pooled cell axis.

qc_histogram: bool

Print the per-cell nnz histogram + the suggested/applied cutoff (the same ASCII summary as data-beans squeeze --show-histogram).

Trait Implementations

impl Clone for QcConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for QcConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for QcConfig

fn default() -> Self

Returns the “default value” for a type.