Skip to contents

Find genes with a consistent directional effect across contrasts

Source: R/find_consistent_hit.R
find_consistent_hit.Rd

Runs the same gene-level method for every contrast of one design matrix, then tests for an effect in all contrasts with combine_consistent_hits.

Usage

find_consistent_hit(
  screenR_Object,
  matrix_model,
  contrasts,
  method = c("fry", "roast", "camera", "zscore"),
  direction = c("Down", "Up"),
  fdr = 0.05,
  fdr_method = c("BH", "BY"),
  method_args = list()
)

Arguments

screenR_Object

A normalized ScreenR object.

matrix_model

A design matrix with rows in the sample order of the object, with sample identifiers as row names. Mismatched names or order are errors (see create_design_matrix). Use a design appropriate for the experimental dependence.

contrasts

A uniquely named list of at least two contrasts accepted by find_roast_hit, such as treatment versus control at each time point. The same design and normalization are used throughout.

method

"fry" (default), "roast", "camera", or "zscore". The zscore method first uses compute_barcode_statistics to obtain replicate-aware barcode Z-scores, then runs find_gene_zscore_hit. Both reference = "random" (its default) and reference = "empirical" are supported through method_args.

direction

The prespecified direction, "Down" or "Up".

fdr

Maximum combined FDR to return; 1 returns all tested genes.

fdr_method

"BH" or "BY" for the combined test across genes.

method_args

A named list of additional arguments for the selected method, for example list(number_barcode = 3), list(nrot = 9999, seed = 42) for ROAST, or list(reference = "empirical", min_barcodes = 4) for Z-score. Arguments controlling the object, model, contrast, direction or filtering cannot be overridden. Z-score uses score = "Zscore".

Value

The tibble described in combine_consistent_hits.

Details

FRY and ROAST test a self-contained effect in every contrast; CAMERA tests enrichment relative to the other genes in every contrast. The Z-score interpretation depends on its chosen reference and its calibration. These are different hypotheses. Legacy barcode-count rules have no gene-level p-values and are not supported.

All tested genes, including opposite observed directions, are retained for combination. The existing single-contrast methods keep their usual defaults. Different gene universes across contrasts are an error. For precomputed complete p-value tables, use combine_consistent_hits().

See also

find_roast_hit, find_camera_hit, find_gene_zscore_hit, find_time_hit

Examples

object <- get0("object", envir = asNamespace("ScreenR"))
design <- create_design_matrix(object)
contrasts <- list(
    Treatment = create_contrast(design, "Treated", "Control"),
    Baseline = create_contrast(design, "Treated", "T1/T2")
)
find_consistent_hit(object, design, contrasts, method = "fry")
#> Warning: 3rows with all zero counts
#> Warning: 3rows with all zero counts
#> # A tibble: 0 × 9
#> # ℹ 9 variables: Gene <chr>, Direction <chr>, NContrasts <int>, PValue <dbl>,
#> #   FDR <dbl>, PValue_Treatment <dbl>, Log2FC_Treatment <dbl>,
#> #   PValue_Baseline <dbl>, Log2FC_Baseline <dbl>