apply_fits — apply_fits • gDRcore

Apply multiple fit functions in a single pass over each BumpyMatrix cell, writing each function's results into its own named output assay.

Usage

apply_fits(
  se,
  fit_fns,
  data_type,
  slicing_cols = NULL,
  slicing_values = NULL,
  input_assay = NULL,
  merge = "merge",
  on_error = "warn",
  fit_source
)

apply_custom_fits(...)

Arguments

se: SummarizedExperiment
fit_fns: named list of functions. Each name becomes an output assay name; each value is a function(data.table) → named list (or a named list of named lists for the shared pre-computation pattern).
data_type: one of "single-agent", "combination", "time-course".
slicing_cols: character vector; NULL uses profile default.
slicing_values: character vector; NULL uses profile default.
input_assay: string; NULL uses profile default.
merge: "merge" or "replace".
on_error: "warn" or "stop".
fit_source: character string stamped as fit_source in every output row.

Value

updated SummarizedExperiment with one new (or updated) assay per entry in fit_fns.

Details

apply_fits() is the performance-efficient alternative to chaining multiple apply_fit calls when two or more fit functions operate on the same input assay. Instead of unsplitting the BumpyMatrix K times (once per function), it traverses each cell once and applies all functions in that single pass.

Use this when:

You have two or more independent fit functions on the same data (e.g. Bliss + HSS on combination data).
A single fit function produces results for multiple output assays (shared pre-computation pattern — see below).

Independent fit functions (most common)

Names of fit_fns become the output assay names:


  apply_fits(
    combo_se,
    fit_fns = list(
      custom_bliss = bliss_fit_fn,
      custom_hss   = hss_fit_fn
    ),
    data_type  = "combination",
    fit_source = "synergy"
  )

Shared pre-computation (advanced)

A single function can return a named list of named lists to write multiple assays while computing expensive intermediates only once:


  bliss_and_hss <- function(dt) {
    sa_curves <- fit_hill_curves(dt)  # expensive — done once
    list(
      custom_bliss = list(bliss_score = compute_bliss(sa_curves, dt)),
      custom_hss   = list(hss_score   = compute_hss(sa_curves, dt))
    )
  }
  apply_fits(
    combo_se,
    fit_fns    = list(bliss_and_hss = bliss_and_hss),
    output_assay_map = c(bliss_and_hss = NA),  # ignored; keys from return value
    ...
  )

The multi-output pattern is detected automatically when a fit function returns a named list whose values are themselves named lists. Each top-level name maps to an assay; the inner named list provides the row columns.

Examples

mae <- gDRutils::get_synthetic_data("finalMAE_small.qs2")
se <- mae[["single-agent"]]
fn_a <- function(dt) list(x_mean = mean(dt$x, na.rm = TRUE))
fn_b <- function(dt) list(x_sd = sd(dt$x, na.rm = TRUE))
se_out <- apply_fits(
  se,
  fit_fns = list(mean_metrics = fn_a, sd_metrics = fn_b),
  data_type = "single-agent",
  fit_source = "demo"
)