Skip to contents

Impute missing values based on spatial coordinates

impute_spe.Rd

impute_spe() carries out imputation for missing data in the primary assay using a specified method from a range of methods. Returns additional assay in the same SPE

Usage

impute_spe(
  spe,
  assay_name = NULL,
  method = NULL,
  group_colname,
  k = NULL,
  protein_missingness = NULL
)

Arguments

spe

SPE containing data to be imputed.

assay_name

name of assay with data to be imputed

method

Method of imputation to be used. See details.

group_colname

Column name in metadata that specifies the group information to use for group_mean or knn_group. Example: ROI_abbreviation.

k

K value to be used for k-nearest neighbor imputation

protein_missingness

Proportion of samples allowed to have missing data for a protein in the given imputation method. Example: When method="global_mean," an protein_missingness of 0.5 indicates that any protein missing data for more than 50% of samples across the entire spatial tissue covered by all samples will be excluded from the imputation method algorithm, and that protein's missing values will not be imputed. When method="group_mean," then protein_missingness of 0.5 indicates that a protein must have data for at least 50% of samples in the specified group to be used in the imputation algorithm and to be imputed.

Value

An SPE with an 'imputed' assay with the appropriate imputation called

Details

Methods options and descriptions:

  • zero: replace missing values with 0

  • median : replace missing values with global median per protein

  • median_half : replace missing values with 1/2 global median per protein

  • mean: replace missing values with global mean per protein

  • group_mean: replace missing values with mean per group, e.g. group (example: ROI) for each protein

  • knn: imputation based on k-nearest neighbors, with proteins as neighbors, based on data from all samples across all groups. NOTE: There will still be NA values if the protein is not expressed in this group.

  • group_knn: imputation based on k-nearest neighbors, with proteins as neighbors, based on data from specified group (e.g. ROI, tissue). NOTE: There will still be NA values if the protein is not expressed in this group.

  • spatial_knn: imputation based on k-nearest neighors in space

Examples


data(pancMeta)
data(protMeta)
data(smallPancData)
# We can put all samples into the same object (for statistical power)
pooledData <- dplyr::bind_cols(smallPancData)
pooled.panc.spe <- convert_to_spe(pooledData,
  pancMeta,
  protMeta,
  feature_meta_colname = "pancProts",
  sample_id = ""
)
#> Spatial object created without spatial coordinate 
#>          column names provided. Distance based analysis will not be enabled.
#> Note: Only mapping metadata for 2986 features out of 3000 data points
# we can try two imputation methods and compare the difference
res <- impute_spe(pooled.panc.spe, method = "mean")
res2 <- impute_spe(pooled.panc.spe, method = "group_mean", 
                  group_colname = "Image")
mean(assay(res, "imputed") - assay(res2, "imputed"), na.rm = TRUE)
#> [1] 0.003118943