Population cuts to take from the other samples

Two cases, decided per panel and marker by how many samples placed their cut at a density minimum (valley_*, otsu):

Usage

peer_cut_replacements(thr, z_max = 3.5, min_peers = 3L, min_peer_frac = 0.5)

Arguments

thr

Data frame with sample_id, panel, marker, threshold, source, one row per sample and marker.

z_max

Robust z beyond which a tail cut is replaced when the marker has no majority of density minima.

min_peers

Minimum number of reference cuts.

min_peer_frac

Minimum share of the panel's samples the reference must cover.

Value

A data frame of replacements (sample_id, panel, marker, old_threshold, old_source, threshold, n_peers, robust_z, reason), or NULL when nothing is replaced.

Details

• At least min_peers and at least min_peer_frac of the samples did. The marker is then bimodal in this assay, and a sample without a minimum – a quantile_fallback or a tailgate* cut – takes the median of the minima. On the KMT2C/KMT2D cohort one sample's CD19 tail cut at 3.24, against minima between 2.28 and 2.50 elsewhere, counted 0.9% B cells where manual gating counted 5.7%.

• Fewer did. The other samples' cuts, fallbacks excluded, are the reference: a fallback takes their median, and a tail cut more than z_max robust SDs from them (MAD floored at 0.25, as for scale_outlier) does too.

Cuts from a control, a config entry or a manual override are never touched.