Rank markers by how differently they are distributed in targets and the rest

WHY THE (median, p1, p99) TRIPLE AND NOT A MEAN DIFFERENCE: a marker can separate a population by having a different SPREAD rather than a different centre – a subset that is uniformly dim on a marker everything else is bimodal on, for instance – and a difference of means scores that at zero. Comparing three order statistics catches shifts in location and in both tails, and costs one pass over the column.

Usage

rank_gate_markers(X, y)

Arguments

X

n x m matrix of marker values

y

0/1 target indicator

Value

data.frame of marker and score, best first

Details

Values are scaled to the unit interval first, so a marker on a wider axis cannot outrank a more informative one purely by having larger numbers.