Skip to contents

apply_combo_scores

Source: R/fit_utils.R
apply_combo_scores.Rd

Compute Bliss and HSA synergy scores for a combination SE, replicating the exact scoring logic of fit_SE.combinations.

Usage

apply_combo_scores(
  se,
  scores_assay = "scores",
  averaged_assay = "Averaged",
  metrics_assay = "Metrics",
  excess_assay = NULL,
  normalization_types = c("GR", "RV"),
  fit_source = "gDR",
  score_FUN = calculate_score,
  isobolograms_assay = "isobolograms"
)

Arguments

se

SummarizedExperiment with "Averaged" and "Metrics" assays (combination experiment).

scores_assay

string; name of the output assay to write scores into. Default "scores".

averaged_assay

string; name of the input assay. Default "Averaged".

metrics_assay

string; name of the assay containing SA fit parameters. Default "Metrics".

excess_assay

string or NULL; if the SE already contains a pre-computed excess assay (e.g. from apply_combo_excess), pass its name here to skip excess re-computation and score directly from it. Default NULL (excess is computed internally).

normalization_types

character vector of normalization types to process. Default c("GR", "RV").

fit_source

string recorded in the fit_source column. Default "gDR".

score_FUN

function to reduce per-point excess values to a scalar score. Default calculate_score (mean of top 10-percentile).

isobolograms_assay

string or NULL; name of the assay with the Loewe isobolograms used for CIScore_50 and CIScore_80. Default "isobolograms".

Value

Updated SummarizedExperiment with a scores_assay assay containing bliss_score, hsa_score, CIScore_50 and CIScore_80 per triplet.

Details

Unlike the low-level bliss_fit_fn and hss_fit_fn (which work on raw Averaged data per triplet), this function uses the SA fit parameters from the Metrics assay to generate curve-smoothed single-agent responses via predict_efficacy_from_conc before computing excess. This produces results numerically identical to fit_SE.combinations.

The function requires a Metrics assay containing columns dilution_drug, cotrt_value, ec50, h, x_inf, x_0, and normalization_type — as produced by fit_SE.combinations or apply_fit_to_se with fit_drug_response_metrics.

Scoring steps (per drug-combo × cell-line × normalization_type):

  1. Predict smooth SA responses at every combo concentration using predict_efficacy_from_conc on drug_1 and drug_2 parameter sets from Metrics.

  2. Average col_values (drug-1-along-conc1) and row_values (drug-2-along-conc2) → smooth.

  3. Compute Bliss-expected using calculate_Bliss and HSA-expected using calculate_HSA on the smooth SA edges.

  4. Compute excess via calculate_excess.

  5. Score = mean of top 10-percentile excess via calculate_score.

CIScore_50 and CIScore_80 are read from the isobolograms_assay written by apply_combo_isobolograms: the combination index of the isobologram at the lowest response level \(\ge\) 0.5 and \(\ge\) 0.2. They are NA when that assay is missing or has no isobologram for the drug pair and cell line.

See also

bliss_fit_fn, hss_fit_fn for the simplified raw-data variants.

Examples

mae <- gDRutils::get_synthetic_data("finalMAE_combo_matrix_small")
combo_se <- mae[[gDRutils::get_supported_experiments("combo")]]
combo_se_out <- apply_combo_scores(combo_se)
"scores" %in% SummarizedExperiment::assayNames(combo_se_out)
#> [1] TRUE