In broadinstitute/inferCNV: Infer Copy Number Variation from Single-Cell RNA-Seq Data

knitr::opts_chunk$set(echo = TRUE)
library(infercnv)

Create the InferCNV Object

Reading in the raw counts matrix and meta data, populating the infercnv object

infercnv_obj = CreateInfercnvObject(
  raw_counts_matrix="oligodendroglioma_expression_downsampled.counts.matrix",
  annotations_file="oligodendroglioma_annotations_downsampled.txt",
  delim="\t",
  gene_order_file="gencode_downsampled.txt",
  ref_group_names=c("Microglia/Macrophage","Oligodendrocytes (non-malignant)"))

Filtering genes

Removing those genes that are very lowly expressed or present in very few cells

# filter out low expressed genes
cutoff=1
infercnv_obj <- require_above_min_mean_expr_cutoff(infercnv_obj, cutoff)

# filter out bad cells
min_cells_per_gene=3
infercnv_obj <- require_above_min_cells_ref(infercnv_obj, min_cells_per_gene=min_cells_per_gene)

## for safe keeping
infercnv_orig_filtered = infercnv_obj
#plot_mean_chr_expr_lineplot(infercnv_obj)
save('infercnv_obj', file = 'infercnv_obj.orig_filtered')

Normalize each cell's counts for sequencing depth

Perform a total sum normalization. Generates counts-per-million or counts-per-100k, depending on the overall sequencing depth.

infercnv_obj <- infercnv:::normalize_counts_by_seq_depth(infercnv_obj)

Spike in artificial variation for tracking purposes

Add ~0x and 2x variation to an artificial spike-in data set based on the normal cells so we can track and later scale residual expression data to this level of variation.

infercnv_obj <- spike_in_variation_chrs(infercnv_obj)

perform Anscombe normalization

<<<<<<< HEAD Suggested for removing noisy variation at low counts ======= Useful noise reduction method.
See: https://en.wikipedia.org/wiki/Anscombe_transform

29a0b973d2701fe5ea2834efcd6a82dd542e0308

infercnv_obj <- infercnv:::anscombe_transform(infercnv_obj)
save('infercnv_obj', file='infercnv_obj.anscombe')

log transform the normalized counts:

infercnv_obj <- log2xplus1(infercnv_obj)
save('infercnv_obj', file='infercnv_obj.log_transformed')

Apply maximum bounds to the expression data to reduce outlier effects

Here we define a threshold by taking the mean of the bounds of expression data across all cells. This is then use to define a cap for the bounds of all data.

threshold = mean(abs(get_average_bounds(infercnv_obj))) 
infercnv_obj <- apply_max_threshold_bounds(infercnv_obj, threshold=threshold)

Initial view, before inferCNV operations:

plot_cnv(infercnv_obj, 
         output_filename='infercnv.logtransf', 
         x.range="auto", 
         title = "Before InferCNV (filtered & log2 transformed)", 
         color_safe_pal = FALSE, 
         x.center = mean([e-mail]))

knitr::include_graphics("infercnv.logtransf.png")

perform smoothing across chromosomes

The expression values are

infercnv_obj = smooth_by_chromosome(infercnv_obj, window_length=101, smooth_ends=TRUE)
save('infercnv_obj', file='infercnv_obj.smooth_by_chr')

# re-center each cell
infercnv_obj <- center_cell_expr_across_chromosome(infercnv_obj, method = "median")
save('infercnv_obj', file='infercnv_obj.cells_recentered')

plot_cnv(infercnv_obj, 
         output_filename='infercnv.chr_smoothed', 
         x.range="auto", 
         title = "chr smoothed and cells re-centered", 
         color_safe_pal = FALSE)

knitr::include_graphics("infercnv.chr_smoothed.png")

subtract the reference values from observations, now have log(fold change) values

infercnv_obj <- subtract_ref_expr_from_obs(infercnv_obj, inv_log=TRUE)
save('infercnv_obj', file='infercnv_obj.ref_subtracted')

plot_cnv(infercnv_obj, 
         output_filename='infercnv.ref_subtracted', 
         x.range="auto", 
         title="ref subtracted", 
         color_safe_pal = FALSE)

knitr::include_graphics("infercnv.ref_subtracted.png")

invert log values

Converting the log(FC) values to regular fold change values, centered at 1 (no fold change)

This is important because we want (1/2)x to be symmetrical to 1.5x, representing loss/gain of one chromosome region.

infercnv_obj <- invert_log2(infercnv_obj)
save('infercnv_obj', file='infercnv_obj.inverted_log')

plot_cnv(infercnv_obj, 
         output_filename='infercnv.inverted', 
         color_safe_pal = FALSE, 
         x.range="auto", 
         x.center=1, 
         title = "inverted log FC to FC")

knitr::include_graphics("infercnv.inverted.png")

Removing noise

infercnv_obj <- clear_noise_via_ref_mean_sd(infercnv_obj, sd_amplifier = 1.0)
save('infercnv_obj', file='infercnv_obj.denoised')

plot_cnv(infercnv_obj, 
         output_filename='infercnv.denoised', 
         x.range="auto", 
         x.center=1, 
         title="denoised", 
         color_safe_pal = FALSE)

knitr::include_graphics("infercnv.denoised.png")

Remove outlier data points

This generally improves on the visualization

infercnv_obj = remove_outliers_norm(infercnv_obj)
save('infercnv_obj', file="infercnv_obj.outliers_removed")

plot_cnv(infercnv_obj, 
         output_filename='infercnv.outliers_removed', 
         color_safe_pal = FALSE, 
         x.range="auto", 
         x.center=1, 
         title = "outliers removed")

knitr::include_graphics("infercnv.outliers_removed.png")

Scale residual expression values according to the Spike-in

Perform rescaling of the data according to the spike-in w/ preset variation levels. Then, remove the spike-in data.

# rescale
infercnv_obj <- scale_cnv_by_spike(infercnv_obj)
# remove the spike-in
infercnv_obj <- remove_spike(infercnv_obj)

Mask out those genes that are not signficantly different from the normal cells

Runs a Wilcoxon rank test comparing tumor/normal for each patient and normal sample, and masks out those genes that are not significantly DE.

infercnv_obj <- infercnv:::mask_non_DE_genes_basic(infercnv_obj, test.use = 't', center_val=1)

save('infercnv_obj', file="infercnv_obj.non_DE_masked")

infercnv_obj <- infercnv:::mask_non_DE_genes_basic(infercnv_obj, center_val=1)

plot_cnv(infercnv_obj, 
         output_filename='infercnv.non-DE-genes-masked', 
         color_safe_pal = FALSE, 
         x.range=c(0,2), # want 0-2 post scaling by the spike-in 
         x.center=1, 
         title = "non-DE-genes-masked")

knitr::include_graphics("infercnv.non-DE-genes-masked.png")

Brighten it up by changing the scale threshold to our liking:

plot_cnv(infercnv_obj, 
         output_filename='infercnv.finalized_view', 
         color_safe_pal = FALSE, 
         x.range=c(0.7, 1.3), 
         x.center=1, 
         title = "InferCNV")

knitr::include_graphics("infercnv.finalized_view.png")

And that's it. You can experiment with each step to fine-tune your data exploration. See the documentation for uploading the resulting data matrix into the Next Generation Clustered Heatmap Viewer for more interactive exploration of the infercnv-processed data: https://github.com/broadinstitute/inferCNV/wiki/Next-Generation-Clustered-Heat-Map

broadinstitute/inferCNV documentation built on Jan. 3, 2024, 6:32 p.m.