Skip to contents

Feature Data

Source: vignettes/feature_data.Rmd
feature_data.Rmd

Feature Counts and Metadata

A feature table is a required file for the qSIP2 pipeline. It is a typical ASV/OTU table where individual taxa are in rows, and sample names are in columns. The table is populated with raw sequencing counts from an amplicon workflow, or some other proxy for abundance (like mean/median depth of coverage) if working with MAGs, contigs or other data types.

A “feature” refers to the names of your individual sequenced units (amplicons, taxa, vOTUs, MAGs, etc.) The feature data should be in a dataframe with a column designated as the feature_id. If you have a dataframe with rownames you can convert that to a column using the tibble::rownames_to_column() function.

df_with_rownames <- data.frame(
  row.names = c("feature1", "feature2", "feature3"),
  sample1 = c(1, 2, 3),
  sample2 = c(4, 5, 6)
)

# data has rownames
rownames(df_with_rownames)
#> [1] "feature1" "feature2" "feature3"
df_with_rownames
#>          sample1 sample2
#> feature1       1       4
#> feature2       2       5
#> feature3       3       6

# convert rownames to their own column
df_with_rownames |>
  tibble::rownames_to_column(var = "feature_id")
#>   feature_id sample1 sample2
#> 1   feature1       1       4
#> 2   feature2       2       5
#> 3   feature3       3       6

Each row corresponds to a feature_id, and the abundance of that feature in a certain sample lives in a column with that sample_id as the column header. The abundance values themselves can be one of several types, and the values are subject to different validation requirements based on the type of the data. Currently, the accepted types are counts, coverage, normalized, and relative.

  • counts is the default and what should be used in most cases where you are giving raw sequencing counts. Here, the data is expected to be integers equal to or greater than 0.
  • coverage is designed for use with MAGs or other data types where you are using a proxy for abundance like mean/median depth of coverage. Here, the data is expected to be numeric values equal to or greater than 0.
  • relative is for situations where you might have lost the original integer count data and only have relative abundance. If using this option it expects fractional abundances (rather than percentages) so each column must sum to 1 or less.
  • normalized is a special case where you have already pre-normalized your counts using an internal spike-in or similar.

qSIP2 Feature Data Object

The qsip_feature_data() function creates the qsip_feature_data object. A qSIP_feature_data object holds validated abundance data for your features. It can be made by giving an already made dataframe, or by modifying a dataframe and piping directly into the function. There is an example dataframe in the qSIP2 package called example_feature_data.

feature_data = qsip_feature_data(example_feature_df, 
                                 feature_id = "ASV",
                                 type = "counts")

Structure of qsip_feature_data

Like the other qSIP2 objects, the qsip_feature_data object contains a @data slot to hold the feature table, but it isn’t intended to be worked with directly. The @type slot holds the type of data, and the @feature_id slot holds the name of the column with the feature ids. There is an additional slot for the taxonomy data, if you have it (see below).

You can return the original dataframe with the get_dataframe() method.

# not run
get_dataframe(feature_data)

Validation of qsip_feature_data

Most of the validation checks depend on the chosen type. If you try to pass values that don’t match the type you specified, you will get an error. For example, fractional values are not allowed when the type is the default counts.

tibble(
  feature_id = c("feature1", "feature2", "feature3"),
  sample1 = c(0.1, 0.2, 0.3),
  sample2 = c(0.4, 0.5, 0.6)
) |>
  qsip_feature_data()
#> Error: Some data are not integers

But it is allowed with type = "coverage".

tibble(
  feature_id = c("feature1", "feature2", "feature3"),
  sample1 = c(0.1, 0.2, 0.3),
  sample2 = c(0.4, 0.5, 0.6)
) |>
  qsip_feature_data(type = "coverage")
#> <qsip_feature_data>
#> feature_id count: 3
#> sample_id count: 2
#> data type: coverage
#> taxonomy: FALSE

In theory, the slots can be overwritten, but this is not recommended. If you do, they will undergo the same validations and may fail.

feature_data@type <- "relative"
#> Error: Some columns have a total relative abundance sum greater than 1

Special Considerations

NA Values

If you have NA values in your abundance table, you will get an error when trying to make the feature object. In most cases a value of NA means an abundance of 0, and so the best practice would be to convert prior to creating the object with a mutate call and the across function from the tidyr package.

tibble(
  feature_id = c("feature1", "feature2", "feature3"),
  sample1 = c(1, 2, NA),
  sample2 = c(4, 5, 6)
) |>
  mutate(across(everything(), ~ replace_na(.x, 0))) |>
  qsip_feature_data()
#> <qsip_feature_data>
#> feature_id count: 3
#> sample_id count: 2
#> data type: counts
#> taxonomy: FALSE

Taxonomy or other metadata

If you have further metadata for your features, such as a taxonomy table, you can add it with the add_taxonomy() function and it will live in the @taxonomy slot.

taxonomy <- tibble(
  feature_id = c("feature1", "feature2", "feature3"),
  genus = c("Marinobacter", "Devosia", "Pseudomonas"),
  species = c("adhaerens", "insulae", "syringae")
)

tibble(
  feature_id = c("feature1", "feature2", "feature3"),
  sample1 = c(1, 2, 3),
  sample2 = c(4, 5, 6)
) |>
  qsip_feature_data() |>
  add_taxonomy(taxonomy, feature_id = "feature_id")
#> <qsip_feature_data>
#> feature_id count: 3
#> sample_id count: 2
#> data type: counts
#> taxonomy: TRUE