Compute variant-to-tree precision/recall/F1 curve across confidence cutoffs

Construct a variant VAF matrix from long-format mutation counts, compare variant partitions against tree clades using Jaccard similarity, and sweep node-confidence cutoffs to summarize precision, recall, and F1.

Usage

compute_variant_pr_curve(
  phy_annot,
  mut_dat,
  min_vaf = 0.01,
  min_cells = 2,
  j_thres = 0.66,
  n_points = 50,
  ncores = 1
)

Arguments

phy_annot

A rooted phylo object with clade confidence stored in node.label.

mut_dat

Long-format mutation table containing at least columns cell, variant, a, and d.

min_vaf

Minimum VAF used when defining variant-positive cells (default 0.01).

min_cells

Minimum number of variant-positive cells required for a variant to be evaluated (default 2).

j_thres

Jaccard threshold for declaring a match between a variant split and a tree clade (default 0.66).

n_points

Number of confidence cutoffs to evaluate (default 50).

ncores

Number of cores for sweeping confidence cutoffs. Uses parallel::mclapply when ncores > 1 on non-Windows systems.

Value

A data frame with columns conf_cutoff, n_pred, n_tp, n_tru, n_recover, precision, recall, f1, and fraction_well_matched.