Struct ReadSharedRowsArgs 

pub struct ReadSharedRowsArgs {

    pub data_files: Vec<Box<str>>,
    pub batch_files: Option<Vec<Box<str>>>,
    pub preload: bool,
    pub feature_kind: Option<FeatureNameKind>,
    pub row_alignment: RowAlignment,
    pub column_alignment: ColumnAlignment,
    pub qc: Option<QcConfig>,
    pub keep_empty_barcodes: bool,
    pub qc_exempt_files: Option<Vec<bool>>,
    pub qc_block_size: Option<usize>,
    pub qc_report_out: Option<Box<str>>,
    pub per_file_feature_suffix: Option<Vec<Box<str>>>,
    pub per_file_barcode_suffix: Option<Vec<Option<Box<str>>>>,
}

Arguments for loading multiple sparse data files with shared row names.

Fields

data_files: Vec<Box<str>>

batch_files: Option<Vec<Box<str>>>

preload: bool

feature_kind: Option<FeatureNameKind>

Cross-file row-name canonicalization rule. None = auto-detect via FeatureNameKind::auto_detect once row names are in hand. Some(kind) skips detection and uses the caller’s choice. Default = None (auto).

row_alignment: RowAlignment

How to align row names across input files. Default RowAlignment::Union keeps every row from any backend — the strictly more permissive option that reduces to single-modality semantics when all files share their row set, and supports multi-modal load (e.g. paired RNA + ATAC) when they don’t. Switch to RowAlignment::Intersect for strict “common rows only” behavior.

column_alignment: ColumnAlignment

How to align column (cell / barcode) names across input files. Default ColumnAlignment::Disjoint concatenates cells with @<basename> disambiguation, preserving single-modality semantics. Switch to ColumnAlignment::Union for patchy multi-modal (multiome) load: cells are glued by raw barcode across backends, a cell observed in only one modality contributes triplets only on that modality’s row block, and no @<basename> suffix is added.

qc: Option<QcConfig>

Optional shared cell QC. When Some, non-near-empty MAD outlier cells are dropped from the working set (via mask_columns), the returned batch Vec is filtered in lockstep, junk features are dropped (when feature_min_cells > 0), and the near-empty output keep-mask is returned in output_keep_idx. None (the default) = no QC, i.e. today’s behavior.

keep_empty_barcodes: bool

Opt out of the empty-barcode gate. By default each file’s columns are cell-called on their own nnz distribution (crate::qc::suggest_nnz_cutoff: the trough between the ambient and the cell peak) and a column is dropped when it falls below the cut in every file that observes it — so an unfiltered barcode axis (e.g. ATAC from a fragments file) loses its empty droplets before any projection or collapse, while a cell in two modalities survives on either. A no-op on already-called data (no trough). Set true where every input column must come back, e.g. query cells at inference. Files flagged in qc_exempt_files are never gated.

qc_exempt_files: Option<Vec<bool>>

Per-data_files entry: true exempts that file’s columns from QC — out of the band statistics AND out of every verdict (see qc_from_metrics). For inputs whose columns are not cells (a carried pb_reference, bulk samples): a pseudobulk standing for hundreds of cells is a legitimate depth outlier, and letting it into the MAD band either gets it dropped or — worse, as the mixture grows — recenters the band and guillotines the real cells. None = no exemption. Must match data_files length when Some.

qc_block_size: Option<usize>

Block size for the QC streaming stat passes (None = default).

qc_report_out: Option<Box<str>>

Optional path for a per-cell QC report TSV (None = don’t write).

per_file_feature_suffix: Option<Vec<Box<str>>>

Optional per-file feature-name (row) modality suffix, one entry per data_files entry in order. When Some, backend b’s rows are renamed {canon(row)}/{suffix[b]} so files sharing raw feature names (e.g. spliced/unspliced) stay on separate rows, while the same name + suffix (same modality across samples) still merges. None = today’s behavior (no suffixing). Must match data_files length.

per_file_barcode_suffix: Option<Vec<Option<Box<str>>>>

Optional per-file barcode (column) suffix, one entry per data_files entry in order. Under ColumnAlignment::Union, backend b’s barcodes are tagged {barcode}@{suffix[b]} before the canonical merge, so callers can encode per-cell sample identity: the same barcode in two files merges into one cell only when both carry the same suffix (same sample across modalities), while different samples stay distinct. None (or a per-entry None) = no tag. Must match data_files length. No effect under Disjoint (which disambiguates via @<basename>).

Trait Implementations

impl Default for ReadSharedRowsArgs

fn default() -> Self

Returns the “default value” for a type.