Identify population-specific markers for patch detection

Patch detection can be applied to identify patches of one cell type on another. The prerequisite for this analysis is that the patch-specific markers are known. Patch detection can be improved further by also leveraging receiver-specific markers. This function identifies population-specific markers for the patch and receiver populations, which can then be used for patch detection.

Usage

identify_markers_for_patch_analysis(
  object,
  group_by,
  receiver_population,
  target_population,
  abundance_difference = 10,
  min_freq = 0.01,
  show_plot = TRUE,
  seed = 123
)

Arguments

object

A Seurat object

group_by

A string specifying the metadata column to group by

receiver_population

A string specifying the receiver population name present in the group_by column

target_population

A string specifying the target population name present in the group_by column

abundance_difference

A numeric value specifying how many times higher the abundance of a marker should be in one population relative to the other. This is only used to label the markers in the output.

min_freq

A numeric value specifying the minimum frequency of a protein to be labeled in the output table.

show_plot

Logical, whether to show a plot summarizing the results.

seed

An integer seed for reproducibility

Value

A tbl_df with the following columns:

• marker: the name of the protein.

• receiver_unmixed_freq: the estimated proportion in the receiver population after unmixing.

• target_unmixed_freq: the estimated proportion in the target population after unmixing.

• receiver_freq: the proportion in the receiver population.

• target_freq: the proportion in the target population.

• label: a label indicating whether the protein is a marker for the receiver or target population. NA values indicate that the protein is unspecific.

Details

Patch detection is sensitive to the selection of markers and therefore requires careful selection. The best markers are both high-abundant and specific to a cell population.

The method requires a Seurat object with a metadata column containing the population information, e.g. a column with cell type labels. We then need to specify the receiver and target populations, where the receiver population represent the cells on which the patches are expected to be found and the target population represents the cell type from which the patches originated.

As a practical example, let's say that our data represent co-cultured T and B cells, and we anticipate that the T cells have patches of B cells on them. Now we face a challenge because the T cell population contains a lot of B cell markers, making it harder to determine what markers are T-cell specific. In other words, the abundance data is mixed.

This method attempts to unmix the abundance data using matrix factorization, estimating the composition of the pure receiver and target populations. The unmixed abundance profiles are then used to label population-specific markers based on the difference in abundance and minimum frequency. The results are summarized in a table, and an optional plot is drawn to help interpret the results.