DIEM

Remove debris-contaminated droplets from single-cell based data.

Installation

Currently, we only support installation of the diem R package through devtools

library(devtools)
devtools::install_github("marcalva/diem")

Usage

Check out the vignette for a thorough tutorial on using diem.

Shown below is a quick workflow for reading 10X data, filtering droplets using default parameters, and converting to a seurat object. Note you need Seurat installed to run the last step.

library(diem)
library(Seurat)

counts <- read_10x("path/to/10x") # Read 10X data into sparse matrix
sce <- create_SCE(counts) # Create SCE object from counts

# Add MT% and MALAT1%
mt_genes <- grep(pattern="^mt-", x=rownames(sce@gene_data), ignore.case=TRUE, value=TRUE)
sce <- get_gene_pct(x = sce, genes=mt_genes, name="pct.mt")
malat <- grep(pattern="^malat1$", x=rownames(sce@gene_data), ignore.case=TRUE, value=TRUE)
sce <- get_gene_pct(x = sce, genes=malat, name="MALAT1")

# Plot total counts ranked
barcode_rank_plot(sce)

# DIEM steps
sce <- set_debris_test_set(sce)
sce <- filter_genes(sce)
sce <- get_pcs(sce)
sce <- init(sce)
sce <- run_em(sce)
sce <- assign_clusters(sce)
sce <- estimate_dbr_score(sce)

# Evaluate debris scores
sm <- summarize_clusters(sce)
plot_clust(sce, feat_x = "n_genes", feat_y = "score.debris", 
           log_x = TRUE, log_y = FALSE)
plot_clust(sce, feat_x = "pct.mt", feat_y = "score.debris", 
           log_x = TRUE, log_y = FALSE)

# Call targets using debris score for single-nucleus data
sce <- call_targets(sce, thresh_score = 0.5)

# Call targets by removing droplets in debris cluster(s) for single-cell data
sce <- call_targets(sce, clusters = "debris", thresh = NULL)

seur <- convert_to_seurat(sce)

Version History

December 19, 2022

• version 2.4.1
  • Fixed an emergent bug that dropped barcode IDs when calling get_pcs and normalizing counts.
  • Fixed a bug in vec_cmplmnt that could have caused invalid indexing. Added further checks in index values in MultEM.cpp.
  • Fixed possible bugs in test functions introduced in v2.4.0

February 13, 2022

• version 2.4.0
  • Added a prior to reduce overfitting

April 13, 2020

• version 2.3.0
  • Quantifies amount of contamination in droplets. Filtering is performed using this debris score.
  • Clustering switched to multinomial mixture model to increase speed.

February 25, 2020

• version 2.2.0
  • Initialize alpha with method of moments instead of optimize

February 19, 2020

• version 2.1.0
  • Additional function for extracting Alpha parameters for use with DE
  • Run multiple k_init values at the same time
  • Multi-threading
  • More efficient memory storage of objects

February 18, 2020

• version 2.0.1
  • Patch fixes some installation issues with tests and docs

February 2, 2020

• version 2.0.0
  • Uses Dirichlet-multinomial
  • Initializes with k-means
  • Removes background centers using a likelihood strategy