In biosurf/cyCombine: Robust integration of single-cell cytometry datasets within and across technologies

knitr::opts_chunk$set(
  strip.white = T, comment = ""
)

This vignette will introduce the panel merging module of cyCombine which has two main functions:

. Imputation of non-overlapping channels for whole data sets (impute_across_panels)

. Imputation of single misstained (or in some other way) problematic channels (salvage_problematic)

For demonstrated use cases and discussions, please refer to the online vignette. Here, we will only introduce the how to use the functions.

Imputation of non-overlapping channels for whole data sets

We start by loading packages, then load the data files using cyCombine's prepare_data function.

# Loading packages
library(cyCombine)
library(tidyverse)

# Setting data directory
data_dir <- 'data_dir'

# Loading two panels' data - this is meant to represent a simple case with only one sample per panel. 
# Refer to the cyCombine reference manual for specifics on how to load more complex sets
panel_1 <- prepare_data(data_dir = data_dir, 
                        pattern = 'panel1', 
                        down_sample = F, 
                        batch_ids = 'Panel1', 
                        sample_ids = 1)

panel_2 <- prepare_data(data_dir = data_dir, 
                        pattern = 'panel2', 
                        down_sample = F, 
                        batch_ids = 'Panel2', 
                        sample_ids = 1)

The data from these two panel can then be merged. We want to specify which markers to base the imputation on (overlapping markers) and which markers to impute (missing_X)

We will do panel merging with two panels at a time. Let us first focus on panels A and B.

# Define the overlap 
overlap_channels <- intersect(get_markers(panel_1), get_markers(panel_2))

# Define markers unique to each panel
missing_1 <- get_markers(panel_2)[!(get_markers(panel_2) %in% overlap_channels)]
missing_2 <- get_markers(panel_1)[!(get_markers(panel_1) %in% overlap_channels)]

# Perform imputations 
imputed <- impute_across_panels(dataset1 = panel_1,
                                dataset2 = panel_2,
                                overlap_channels = overlap_channels,
                                impute_channels1 = missing_1,
                                impute_channels2 = missing_2)

# Extract the dataframes for each set
imputed_1 <- imputed$dataset1
imputed_2 <- imputed$dataset2

# Bind the rows together - good for co-analysis
final <- rbind(imputed$dataset1, imputed$dataset2)

Imputation of single misstained channels

Sometimes we want to correct only a single channel - we will show how to do so here.

# Loading packages 
library(cyCombine)
library(tidyverse)


## Loading data based on a panel and meta-data defined in csv format

# Specifying data directory
data_dir <- 'data_dir'

# Loading panel info
panel <- read_csv(paste0(data_dir, "/panel.csv"))

# Extracting the markers
markers <- panel %>%
  filter(Type != "none") %>%
  pull(Marker) %>%
  str_remove_all("[ _-]")

# Preparing the expression data
data <- prepare_data(data_dir = data_dir,
                     metadata = paste0(data_dir, "/metadata.csv"),
                     filename_col = "FCS_name",
                     batch_ids = "Batch",
                     condition = "Set",
                     derand = TRUE,
                     cofactor = 5,
                     markers = markers,
                     down_sample = FALSE)

Now that we have loaded the data, we want to correct a single misstained channel, which we will refer to as 'BAD'. It was misstained in two batches, batches 3 and 4. It worked fine in the other 4 batches of the data: Batches 1-2 and 5-6.

# Salvage BAD in batches 3-4
fixed <- salvage_problematic(df = data,
                             correct_batches = c(3,4),
                             channel = 'BAD',
                             sample_size = 100000)

We can visualize this correction with density plots. Only 'BAD' will be affected.

plot_density(data, fixed, ncol = 4)

biosurf/cyCombine documentation built on May 23, 2024, 4:07 a.m.