TMSig: Tools for Analyzing Molecular Signatures

The TMSig R package contains tools to prepare, analyze, and visualize named lists of mathematical sets, with an emphasis on molecular signatures (such as gene or kinase sets). It includes fast, memory efficient functions to construct sparse incidence and similarity matrices and filter, cluster, invert, and decompose sets. Additionally, bubble heatmaps can be created to visualize the results of any differential or molecular signatures analysis.

We define a molecular signature as any collection of genes, proteins, post-translational modifications (PTMs), metabolites, lipids, or other molecules with an associated biological interpretation. Most molecular signatures databases are gene-centric, such as the Molecular Signatures Database (MSigDB; Liberzon et al., 2011, 2015), though there are others like the Metabolomics Workbench Reference List of Metabolite Names (RefMet) database (Fahy & Subramaniam, 2020).

Installation

To install this package, start R (>= 4.4.0) and enter:

if (!require("BiocManager", quietly = TRUE))
    install.packages("BiocManager")

BiocManager::install("TMSig")

You can install the development version of TMSig like so:

if (!require("devtools", quietly = TRUE))
  install.packages("devtools")

# Install package and build vignettes
devtools::install_github("EMSL-Computing/TMSig", build_vignettes = TRUE)

Overview

Below is an overview of some of the core functions.

• readGMT: create a named list of sets from a GMT file.

• sparseIncidence: compute a sparse incidence matrix with unique sets as rows and unique elements as columns. A value of 1 indicates that a particular element is a member of that set, while a value of 0 indicates that it is not.

• similarity: compute the matrix of pairwise Jaccard, overlap, or Ōtsuka similarity coefficients for all pairs of sets $AA$ and $BB$, where

  • Jaccard(A, B) = $\frac{\mathrm{\mid }A\cap B\mathrm{\mid }}{\mathrm{\mid }A\cup B\mathrm{\mid }}\backslash frac\{|A\; \backslash cap\; B|\}\{|A\; \backslash cup\; B|\}$
  • Overlap(A, B) = $\frac{\mathrm{\mid }A\cap B\mathrm{\mid }}{\min (\mathrm{\mid }A\mathrm{\mid },\mathrm{\mid }B\mathrm{\mid })}\backslash frac\{|A\; \backslash cap\; B|\}\{\backslash min(|A|,\; |B|)\}$
  • Ōtsuka(A, B) = $\frac{\mathrm{\mid }A\cap B\mathrm{\mid }}{\sqrt{\mathrm{\mid }A\mathrm{\mid }\times \mathrm{\mid }B\mathrm{\mid }}}\backslash frac\{|A\; \backslash cap\; B|\}\{\backslash sqrt\{|A|\; \backslash times\; |B|\}\}$

• filterSets: restrict sets to only those elements in a pre-determined background, if provided, and only keep those that pass minimum and maximum size thresholds.

• clusterSets: hierarchical clustering of highly similar sets. Used to reduce redundancy prior to analysis.

• decomposeSets: decompose all pairs of sufficiently overlapping sets, $AA$ and $BB$, into 3 disjoint parts.

  1. The elements unique to $AA$: $A\setminus BA\; \backslash setminus\; B$ (“A minus B”)
  2. The elements unique to $BB$: $B\setminus AB\; \backslash setminus\; A$ (“B minus A”)
  3. The elements common to both $AA$ and $BB$: $A\cap BA\; \backslash cap\; B$ (“A and B”)

• invertSets: swap positions of sets and elements so that elements become set names and set names become elements.

• cameraPR.matrix: a fast matrix method for limma::cameraPR for testing molecular signatures in one or more contrasts. Pre-Ranked Correlation Adjusted MEan RAnk gene set testing (CAMERA-PR) accounts for inter-gene correlation to control the type I error rate (Wu & Smyth, 2012).

• enrichmap: visualize molecular signature analysis results, such as those from cameraPR.matrix, as a bubble heatmap with signatures as rows and contrasts as columns.

Examples

Please refer to vignette(topic = "TMSig", package = "TMSig") for examples of how to use this package.

library(TMSig)
#> Loading required package: limma

# Named list of sets
x <- list("Set1" = letters[1:5],
          "Set2" = letters[1:4], # subset of Set1
          "Set3" = letters[1:4], # aliased with Set2
          "Set4" = letters[1:3], # subset of Set1-Set3
          "Set5" = c("a", "a", NA), # duplicates and NA
          "Set6" = c("x", "y", "z"), # distinct elements
          "Set7" = letters[3:6]) # overlaps with Set1-Set5
x
#> $Set1
#> [1] "a" "b" "c" "d" "e"
#> 
#> $Set2
#> [1] "a" "b" "c" "d"
#> 
#> $Set3
#> [1] "a" "b" "c" "d"
#> 
#> $Set4
#> [1] "a" "b" "c"
#> 
#> $Set5
#> [1] "a" "a" NA 
#> 
#> $Set6
#> [1] "x" "y" "z"
#> 
#> $Set7
#> [1] "c" "d" "e" "f"

(imat <- sparseIncidence(x)) # incidence matrix
#> 7 x 9 sparse Matrix of class "dgCMatrix"
#>      a b c d e x y z f
#> Set1 1 1 1 1 1 . . . .
#> Set2 1 1 1 1 . . . . .
#> Set3 1 1 1 1 . . . . .
#> Set4 1 1 1 . . . . . .
#> Set5 1 . . . . . . . .
#> Set6 . . . . . 1 1 1 .
#> Set7 . . 1 1 1 . . . 1

tcrossprod(imat) # pairwise intersection and set sizes
#> 7 x 7 sparse Matrix of class "dsCMatrix"
#>      Set1 Set2 Set3 Set4 Set5 Set6 Set7
#> Set1    5    4    4    3    1    .    3
#> Set2    4    4    4    3    1    .    2
#> Set3    4    4    4    3    1    .    2
#> Set4    3    3    3    3    1    .    1
#> Set5    1    1    1    1    1    .    .
#> Set6    .    .    .    .    .    3    .
#> Set7    3    2    2    1    .    .    4

crossprod(imat) # occurrence of each element and pair of elements
#> 9 x 9 sparse Matrix of class "dsCMatrix"
#>   a b c d e x y z f
#> a 5 4 4 3 1 . . . .
#> b 4 4 4 3 1 . . . .
#> c 4 4 5 4 2 . . . 1
#> d 3 3 4 4 2 . . . 1
#> e 1 1 2 2 2 . . . 1
#> x . . . . . 1 1 1 .
#> y . . . . . 1 1 1 .
#> z . . . . . 1 1 1 .
#> f . . 1 1 1 . . . 1

## Calculate matrices of pairwise Jaccard and overlap similarity coefficients
similarity(x) # Jaccard (default)
#> 7 x 7 sparse Matrix of class "dgCMatrix"
#>      Set1      Set2      Set3      Set4      Set5 Set6      Set7
#> Set1  1.0 0.8000000 0.8000000 0.6000000 0.2000000    . 0.5000000
#> Set2  0.8 1.0000000 1.0000000 0.7500000 0.2500000    . 0.3333333
#> Set3  0.8 1.0000000 1.0000000 0.7500000 0.2500000    . 0.3333333
#> Set4  0.6 0.7500000 0.7500000 1.0000000 0.3333333    . 0.1666667
#> Set5  0.2 0.2500000 0.2500000 0.3333333 1.0000000    . .        
#> Set6  .   .         .         .         .            1 .        
#> Set7  0.5 0.3333333 0.3333333 0.1666667 .            . 1.0000000

similarity(x, type = "overlap") # overlap
#> 7 x 7 sparse Matrix of class "dgCMatrix"
#>      Set1 Set2 Set3      Set4 Set5 Set6      Set7
#> Set1 1.00  1.0  1.0 1.0000000    1    . 0.7500000
#> Set2 1.00  1.0  1.0 1.0000000    1    . 0.5000000
#> Set3 1.00  1.0  1.0 1.0000000    1    . 0.5000000
#> Set4 1.00  1.0  1.0 1.0000000    1    . 0.3333333
#> Set5 1.00  1.0  1.0 1.0000000    1    . .        
#> Set6 .     .    .   .            .    1 .        
#> Set7 0.75  0.5  0.5 0.3333333    .    . 1.0000000

similarity(x, type = "otsuka") # Ōtsuka
#> 7 x 7 sparse Matrix of class "dgCMatrix"
#>           Set1      Set2      Set3      Set4      Set5 Set6      Set7
#> Set1 1.0000000 0.8944272 0.8944272 0.7745967 0.4472136    . 0.6708204
#> Set2 0.8944272 1.0000000 1.0000000 0.8660254 0.5000000    . 0.5000000
#> Set3 0.8944272 1.0000000 1.0000000 0.8660254 0.5000000    . 0.5000000
#> Set4 0.7745967 0.8660254 0.8660254 1.0000000 0.5773503    . 0.2886751
#> Set5 0.4472136 0.5000000 0.5000000 0.5773503 1.0000000    . .        
#> Set6 .         .         .         .         .            1 .        
#> Set7 0.6708204 0.5000000 0.5000000 0.2886751 .            . 1.0000000

## Cluster sets based on their similarity

# Cluster aliased sets
clusterSets(x, cutoff = 1)
#>    set cluster set_size
#> 1 Set2       1        4
#> 2 Set3       1        4
#> 3 Set1       2        5
#> 4 Set4       3        3
#> 5 Set5       4        1
#> 6 Set6       5        3
#> 7 Set7       6        4

# Cluster subsets
clusterSets(x, cutoff = 1, type = "overlap")
#>    set cluster set_size
#> 1 Set1       1        5
#> 2 Set2       1        4
#> 3 Set3       1        4
#> 4 Set4       1        3
#> 5 Set5       1        1
#> 6 Set6       2        3
#> 7 Set7       3        4

Issues

If you encounter a problem with TMSig, please create a new issue that includes:

1. A clear statement of the problem in the title
2. A (small) reproducible example
3. Additional detailed explanation, as needed
4. Output of sessionInfo()

Pull Requests

• Verify that devtools::check(document = TRUE) runs without errors, warnings, or notes before submitting a pull request.
• All contributed code should, ideally, adhere to the tidyverse style guide: https://style.tidyverse.org/index.html. This makes it easier for others to understand, diagnose problems, and make changes. When in doubt, refer to the existing codebase.

References

Fahy, E., & Subramaniam, S. (2020). RefMet: A reference nomenclature for metabolomics. Nature Methods, 17(12), 1173–1174. doi:10.1038/s41592-020-01009-y

Liberzon, A., Subramanian, A., Pinchback, R., Thorvaldsdóttir, H., Tamayo, P., & Mesirov, J. P. (2011). Molecular signatures database (MSigDB) 3.0. Bioinformatics, 27(12), 1739–1740. doi:10.1093/bioinformatics/btr260

Liberzon, A., Birger, C., Thorvaldsdóttir, H., Ghandi, M., Mesirov, J. P., & Tamayo, P. (2015). The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell systems, 1(6), 417–425. doi:10.1016/j.cels.2015.12.004

Wu, D., & Smyth, G. K. (2012). Camera: A competitive gene set test accounting for inter-gene correlation. Nucleic Acids Research, 40(17), e133–e133. https://doi.org/10.1093/nar/gks461