Skip to contents

Put the quantification uncertainty into the counts

Source: R/uncertainty.R
inferential-overdispersion.Rd

Reads that map to several isoforms make the counts of an isoform more variable than a Poisson count, by an amount that differs between transcripts. Baldoni et al. (2024) estimate this overdispersion for every transcript from the Salmon replicates and divide it out of the counts, so that tests built for counts (edgeR, limma, DESeq2) keep their error rate without a model of their own for the read ambiguity.

  • inferential_overdispersion() computes the overdispersion of every transcript from the replicates stored in the object, as edgeR::catchSalmon() does: the Pearson statistic of a quasi-Poisson fit to the replicates of each sample, pooled over the samples, moderated towards the median with a prior of prior_df degrees of freedom and floored at one. The estimates use the transcripts whose replicates are in the object (those with a mean count of at least min_count in add_uncertainty()), so the prior differs from the one catchSalmon() computes over every transcript of the index; transcripts without replicates get the prior.

  • scale_counts() divides an assay by the overdispersion and stores the result as a new assay, ready for a count-based test.

The overdispersion complements the confidence classes of de_confidence(): it changes the test, while the classes describe how far each call can be trusted.

Usage

inferential_overdispersion(self, prior_df = 3, ...)

scale_counts(self, assay = "counts", name = paste0(assay, "_scaled"), ...)

Arguments

self

An IsoformicExperiment with inferential replicates (see add_uncertainty()).

prior_df

Degrees of freedom of the prior (3 in edgeR).

...

Unused.

assay

Name of the counts assay to scale.

name

Name of the new assay.

Value

inferential_overdispersion(): a tibble with transcript_id, overdispersion and df (degrees of freedom of the estimate), with the prior in the prior attribute. scale_counts(): the object with the scaled assay added.

References

Baldoni PL, Chen Y, Hediyeh-Zadeh S, Liao Y, Dong X, Ritchie ME, Shi W, Smyth GK (2024). Dividing out quantification uncertainty allows efficient assessment of differential transcript expression with edgeR. Nucleic Acids Research 52(3):e13.

See also

add_uncertainty(), de_confidence()

Examples

extdata <- system.file("extdata", package = "isoformic")
quant_dirs <- list.files(file.path(extdata, "mini_quants"), full.names = TRUE)
names(quant_dirs) <- sub("_quant$", "", basename(quant_dirs))
iso <- read_salmon(
  quant_dirs,
  annot_path = file.path(extdata, "mini_annotation"),
  col_data = data.frame(
    sample_id = names(quant_dirs),
    condition = rep(c("control", "treatment"), each = 3)
  )
)
iso@metadata$inf_reps <- as_inferential_replicates(read_parquet(
  file.path(extdata, "mini_draws.parquet")
))
iso <- scale_counts(iso)
assay_names(iso)
#> [1] "abundance"     "counts"        "length"        "counts_scaled"
head(inferential_overdispersion(iso))
#> # A tibble: 6 × 3
#>   transcript_id      overdispersion    df
#>   <chr>                       <dbl> <dbl>
#> 1 ENST00000367468.10           2.96   114
#> 2 ENST00000681605.1            5.30   114
#> 3 ENST00000680451.1            3.93   114
#> 4 ENST00000490885.6            1      114
#> 5 ENST00000559627.1            1.58   114
#> 6 ENST00000466691.1            1      114