Compile the weights

Usage

compile_weights(
  weights,
  gs,
  level,
  range = c(-5, -1, 1, 5),
  format = c("universal", "db"),
  drop_zero = TRUE,
  pos_only = FALSE
)

weights: A data frame with the weights
gs: The gene sets
level: The level of the gene sets
range: The quantization range for format = "db" only: the 4-element form c(-low, -high, low, high) (default c(-5, -1, 1, 5)) scales the positive weights into [low, high] and the negative ones into [-high, -low] before rounding them to integers (the db format can only encode integer weights via +/- repeats). The universal output is never rescaled — it keeps the raw trained weights.
format: The format of the output data frame. One of "universal" (default) or "db".
drop_zero: Whether to drop markers whose trained weight is exactly zero from the output (default TRUE). Zero weights mean the marker carries no learned direction for the cell type; dropping them (before any db quantization) makes zero mean "no direction" — the marker is absent from the cell type's list. Pass FALSE to keep every candidate marker.
pos_only: Whether to keep only the markers whose trained weight is positive in the output (default FALSE). A marker trained with a negative weight anti-correlates with the cell type and is reported with direction = "negative"; pass TRUE to drop those rows (as well as any exactly-zero ones, so drop_zero is then redundant) and keep only the markers overexpressed in the cell type.

A data frame with the compiled weights in the universal marker format (default) or the db format (format = "db"), consumable by gs_prepare().