Visualisation Guide • sigvis

library(sigvis)
library(sigstash)
library(TCGAcatalogues)

Warning: this package is in early development and not ready for general use

Background

Mutational signature analysis answers 6 key questions. Each answer can be summarized with a visualization.

What is the observed mutational profile of a single sample
Are there patterns in the observed mutational profile of a cohort?
Which samples are my sample of interest most similar to with respect to their observed mutational profile?
Does mutational signature analysis produce a good quality model from which we can draw robust inferences about a single sample?
What does the mutational signature model of a sample say about a single sample?
Are there patterns in the mutational signature models of a cohort?
What does signature X look like?

Observed mutational profile

1. Single Sample Visualisation


# Load a catalogue
tally <- catalogues_load("BRCA", type = "SBS_96")

# Select a single dataframe of interest
sample = "TCGA-5L-AAT1-01A-12D-A41F-09"
tally_single_sample <- tally[[sample]]

tally_single_sample
#> # A tibble: 96 × 4
#>    channel type  fraction count
#>    <chr>   <chr>    <dbl> <int>
#>  1 A[C>A]A C>A   0.00151      3
#>  2 A[C>A]C C>A   0.00151      3
#>  3 A[C>A]G C>A   0.000503     1
#>  4 A[C>A]T C>A   0            0
#>  5 A[C>G]A C>G   0.00151      3
#>  6 A[C>G]C C>G   0.00151      3
#>  7 A[C>G]G C>G   0            0
#>  8 A[C>G]T C>G   0.00201      4
#>  9 A[C>T]A C>T   0.00402      8
#> 10 A[C>T]C C>T   0.00151      3
#> # ℹ 86 more rows

# Visualise it
sig_visualise(tally_single_sample, class = "catalogue", title = sample)
#> ✔ All channels matched perfectly to set [sbs_96]. Using this set for sort order
#> ✔ All types matched perfectly to set [sbs_type]. Using this set for sort order
#> ✔ Types matched perfectly to palette [snv_type]

2A. Cohort stacked bar

2B. Per channel feature distribution in cohort context (dotplots)

Sample Similarity

signature similarity tree / tsne

Signature Model

Quality

4A. Model vs Observed

4B. Residual Plot

4C. Bootstrap performance

Inferences

Signature Contribution Plot
Signature composition stacked barplot (compatible with ggoncoplot)

Signatures

What does signature X look like?

Often there’ll be a mutational signature and want to visualise the profile.

For example lets plot the profile of signature SBS2 from COSMIC database (but same process will work for de-novo signatures)

# Load Signature
signatures <- sig_load("COSMIC_v3.3.1_SBS_GRCh38")

# Select just SBS2
sbs2 <- signatures[["SBS2"]]

# Plot
sig_visualise(signature = sbs2, title = "SBS2")
#> ✔ All channels matched perfectly to set [sbs_96]. Using this set for sort order
#> ✔ All types matched perfectly to set [sbs_type]. Using this set for sort order
#> ✔ Types matched perfectly to palette [snv_type]

We can do the same with doublets, indels or de novo signatures

# Load Signature
signatures <- sig_load("COSMIC_v3.3_DBS_GRCh38")

# Select just SBS2
dbs9 <- signatures[["DBS9"]]

# Plot
sig_visualise(signature = dbs9, title = "DBS9")
#> ✔ All channels matched perfectly to set [dbs_78]. Using this set for sort order
#> ✔ All types matched perfectly to set [doublet_type]. Using this set for sort order
#> ✔ Types matched perfectly to palette [dbs_type]

Custom Data and Palettes

sig_visualise allows users to provide custom colour mappings and therefore supports all manner of custom visualisations.

# Create Custom Signature
sig_custom <- data.frame(
  type = c("INDELS", "INDELS", "SUBSTITUTIONS","SUBSTITUTIONS"),
  channel = c("INDELS", "DELETIONS", "SBS", "DOUBLETS"),
  fraction = c(0.1, 0.1, 0.7, 0.1)
  )

# Custom palette
 custom_pal <- c('INDELS' = "darkorange", "SUBSTITUTIONS" = "darkgreen")

# Plot
sig_visualise(signature = sig_custom, title = "Custom Signature", palette = custom_pal)
#> ! No exact channel set match found, sorting in order of appearance
#> ! No exact type set match found, sorting in order of appearance