Skip to contents

Compute per-period growth rates for time-course screens

Source: R/fit_SE.timecourse.R
compute_growth_rates.Rd

Stage 1 of the time-course pipeline: fits a growth rate per well within each time window, aggregates replicates, and normalises the rates to the control wells of a reference period. The returned table keeps every treatment, controls included, together with the raw rate and its standard deviation, so it can be reported on directly. Pass it through growth_rates_to_se to obtain the stage 2 input, or use fit_SE.timecourse to run both stages at once.

Usage

compute_growth_rates(
  se,
  periods,
  normalization_map,
  lfc_assay = NULL,
  rate_fn = NULL
)

Arguments

se

SummarizedExperiment with a log-fold-change assay.

periods

named list of numeric vectors of length 2; half-open time windows c(start_hour, end_hour), see fit_SE.timecourse.

normalization_map

named character vector mapping every period to its reference period or to "None", see fit_SE.timecourse.

lfc_assay

string or NULL; name of the log-fold-change assay. NULL uses the input_assay of the "time-course" fit profile (see get_fit_profile).

rate_fn

function or NULL; growth-rate estimator. It receives a data.table holding the timepoints of one well within one window — including the duration column and the log-fold-change column — and must return a single numeric value: the growth rate in doublings per day. NULL uses the slope of lm(LogFoldChange ~ Duration).

Value

A data.table with one row per (CellLine, Drug, Concentration, partner slots, period) — controls included — holding GrowthRate (doublings/day), sd_GrowthRate across replicate wells, rate_0 (the control baseline pooled over all plates, NA for periods mapped to "None" and when no control rows are present), NormalizedGrowthRate with sd_NormalizedGrowthRate across replicate wells, and normalization_type. All of these columns are always present, whatever the normalization map or the availability of controls. NormalizedGrowthRate is NA where the control baseline is not meaningfully positive: such a denominator either inverts the sign of the ratio (when negative) or leaves it undefined (when zero, within fitting tolerance) rather than scaling it, so no meaningful normalized rate exists. A warning names the cell line and period.

Details

Growth rates are computed per (Barcode, WellRow, WellColumn, CellLine, Drug, Concentration, partner drug/concentration slots) group, so single-agent, doublet and triplet arms are never pooled. Control wells are detected via gDRutils::get_env_identifiers("untreated_tag"). Each well is normalized to the mean control rate of its own plate and cell line before the replicates are averaged, so plates that grow at different rates do not widen the spread; a plate without a control of its own falls back to the baseline pooled over all plates, with a warning.

See also

growth_rates_to_se, fit_SE.timecourse, apply_fit

Examples

if (FALSE) { # \dontrun{
periods <- list(early = c(0, 48), late = c(48, 96))
norm_map <- c(early = "None", late = "early")
growth_dt <- compute_growth_rates(se_tc, periods, norm_map)

# Stage 2 with any custom fit function:
se_fit <- gDRutils::apply_fit(
  growth_rates_to_se(growth_dt),
  fit_fn = my_fit_fn,
  data_type = "time-course-metrics",
  output_assay = "Metrics",
  merge = "replace",
  fit_source = "custom"
)
} # }