Skip to contents

One named list shared by all parts of the caller. The calibrated values (shifts, null_q, the depth strata) were set on SPLONGGET normal-tissue and tumour data; see the notes per entry.

Usage

caller_defaults()

Value

named list:

  • null_q threshold quantile of the null score (0.999, type 7: 1 false call per 1,000 null cell-regions); calls need score > threshold (strict, not >=);

  • eps allele error floor of the beta-binomial (0.02);

  • mol_gap fragments of one cell closer than this (bp) are one molecule (2000; the default of cell_features() mol_gap);

  • mol_split_ps split molecules by phase set (FALSE): a molecule whose fragments lie in different phase sets (across a phase-block junction) gives one vote per phase set instead of one vote under the phase set of its first fragment (collapse_molecules() split_ps);

  • votes_clean votes from clean fragments only (FALSE): fragments with the blacklist bit in flags (the fragments the extractor leaves out of the depth layers) do not vote (read_votes() clean);

  • sd_floor floor of the per-cell null SD of the window log2 ratio (0.05);

  • strata_breaks / strata_labels depth strata on clean fragments per cell (<20k, 20-50k, >=50k);

  • shifts log2 depth shifts of loss (-0.98) and gain (+0.57) (calibrated on depth-only spike-ins in normal tissue);

  • k_amp uncalibrated depth shift factor for cn >= 4: shift = k_amp * log2(cn / 2);

  • f_att attenuated major-haplotype fractions of a 1:0 / 2:0 state;

  • tile_sizes named tile sizes (bp);

  • n_folds cross-validation folds of the null thresholds (4);

  • n_pool cells per pool (20);

  • chrx_centromere chrX centromere (bp, hg38) for female chrX arms;

  • acrocentric_p arms without unique sequence, never regions;

  • min_null_rows fewer null rows than this in a class x stratum fall back to the pooled threshold (5000);

  • tabix path of the tabix binary ("" = the tabix on the PATH; none = the R fallback); seed RNG seed (1);

  • calib per-cell calibration of normal mode: "class" (one Student-t null and one threshold per class x stratum, the default design) or "votes" (opt-in: thresholds per class x stratum x vote stratum hstratum, bins of the allele votes n_hap; vote_bins());

  • vote_bins maximal number of n_hap bins per class x stratum of the vote calibration (4; fewer when the class x stratum has fewer than vote_bins x min_null_rows null rows with votes);

  • calib_rho allelic overdispersion of the normal-mode scores: "global" (one rho on the null arm rows) or "chrom" (opt-in: per chromosome, shrunk to the global; rho_by_chrom());

  • rho_chrom_k shrinkage pseudo-count of the per-chromosome rho (100 null arm rows; a chromosome with ~4,000 null rows keeps ~98 % of its own estimate);

  • mask_allelic data-driven allelic tile mask in normal mode (allelic_noisy_tiles(), call_scn()): the votes of 5 Mb tiles whose null overdispersion is an outlier are left out of the allele features (FALSE);

  • mask_rho_mad mask threshold in MADs of log tile rho (2);

  • mask_min_votes minimal mean votes per null row for a tile to be assessed (3);

  • mask_f0 shared-bias allelic tile mask in normal mode (allelic_noisy_tiles(), call_scn()): the votes of 5 Mb tiles whose vote-weighted null-cell haplotype fraction f0 is off 0.5 in an outlying way are left out of the allele features (FALSE; independent of mask_allelic, both on = the union of the two masks; allelic_bias_tiles() shows the per-tile fractions before calling); a real clonal event whose carriers remain among the null cells can be masked too, so read the masked tiles in the run notes; not applied under mirrored orientation (noted and warned);

  • mask_f0_mad bias threshold in MADs of |f0 - 0.5| over the assessed tiles (3);

  • mask_f0_min minimal |f0 - 0.5| of a masked tile (0.03);

  • pool_sex_thr males: the sex chromosomes get their own pooled threshold group (class x autosome / X / Y) with the class threshold as fallback (TRUE; no effect in females);

  • pool_events normal mode: call named event regions (no null rows of their own) also in pools (FALSE). Off, their pools are scored but get no threshold and are never called; their per-cell calls are not affected. On (experimental): their pools are standardised with the class-median scale of their sd_class and thresholded with its pooled threshold, which is not calibrated per event (the null pooled score of a region depends on its own vote count, which the per-region scale of pool_scale() corrects for);

  • pool_tail_q pooled per-region tail calibration (pool_scale()): quantile of the region-standardised null pools compared with the class (0 = off; e.g. 0.99);

  • pool_tail_k shrinkage pseudo-count of the pooled tail factor (30 effective pools);

  • pool_tail_xfit cross-fit the pooled tail (pooled_thresholds()): the region scale and the threshold are estimated on pools of two disjoint halves of the null cells (TRUE).