In insightsengineering/hermes: Preprocessing, analyzing, and reporting of RNA-seq data

The HermesData

The HermesData constructor shall borrow ideas from the legacy se2gSE and gSummarizedExperiment functions. To be specific,

• The gSummarizedExperiment() takes assays, rowData, colData and metadata as input, while HermesData() should directly take SummarizedExperiment object as input, similar like se2gSE().

• The first assay must be counts containing non-missing, integer, non-negative values.

• Similar to the gSummarizedExperiment(), HermesData() must have mandatory rowData (HGNC, HGNCGeneName, etc) and colData (Filename, SampleID, etc) columns. If the input does not include these rowData or colData columns, the function will fill the columns with NA values.

Note: At this point, we don't yet look up gene annotation data. It will be updated later.

Idea: take SE as argument

Let's say we need to have an assay in the SE. Otherwise we don't have any data.

First: what is if there is no colData? Let's try with a very simple SE:

library(SummarizedExperiment)
se <- SummarizedExperiment(matrix(1:4, 2, 2))
se
rownames(colData(se))

OK, if the assay did not have row names, the SE also does not have. But if the assay has row names:

my_assay <- matrix(1:4, 2, 2, dimnames = list(c("a", "b"), c("c", "d")))
rownames(my_assay)
se <- SummarizedExperiment(my_assay)
rownames(colData(se))
rownames(rowData(se))

Then the SE also has row names.

So if we take an SE, it will already have correct row names for colData and rowData. We can see what goes wrong when using the internal constructor:

# hermes:::.HermesData(se)

OK so what we want to help the user is to add the mandatory columns for colData and rowData, as well as annotate the genes (i.e. adding row data) via IGIS (we'll implement that in the next sprint).

Question: there are some columns can't be completely NA. These are captured in the internal package constants .row_data_non_empty_cols and .col_data_non_empty_cols. We don't try to impute these in the constructor since later this would anyway fail the validation. That means these columns need to be supplied by the input object already.

annotate_igis <- identity # to be implemented later

HermesData <- function(object) {
  assert_that(
    is_class(object, "SummarizedExperiment"),
    not_empty(assays(object)),
    not_empty(rowData(object)),
    not_empty(colData(object))
  )

  object <- annotate_igis(object)

  missing_row <- setdiff(.row_data_additional_cols, names(rowData(object)))
  rowData(object)[, missing_row] <- NA

  missing_col <- setdiff(.col_data_additional_cols, names(colData(object)))
  colData(object)[, missing_col] <- NA

  .HermesData(object)
}

We don't tinker with the assays here since that needs to be already correctly set in the object (counts assay with non-negative integers etc.).

insightsengineering/hermes documentation built on Dec. 15, 2024, 8:07 a.m.