mitch

Antony Kaspi & Mark Ziemann

2020-10-09

Background

mitch is an R package for multi-contrast enrichment analysis. At it’s heart, it uses a rank-MANOVA based statistical approach to detect sets of genes that exhibit enrichment in the multidimensional space as compared to the background. The rank-MANOVA concept dates to work by Cox and Mann (https://doi.org/10.1186/1471-2105-13-S16-S12). mitch is useful for pathway analysis of profiling studies with one, two or more contrasts, or in studies with multiple omics profiling, for example proteomic, transcriptomic, epigenomic analysis of the same samples. mitch is perfectly suited for pathway level differential analysis of scRNA-seq data.

The main strengths of mitch are that it can import datasets easily from many upstream tools and has advanced plotting features to visualise these enrichments. mitch consists of five functions. A typical mitch workflow would consist of:

if(!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install("mitch")
library("mitch")
## Registered S3 method overwritten by 'GGally':
##   method from   
##   +.gg   ggplot2

Importing gene sets

mitch has a function to import GMT files to R lists (adapted from Yu et al, 2012 in the clusterProfiler package). For example we can grab some gene sets from Reactome.org:

download.file("https://reactome.org/download/current/ReactomePathways.gmt.zip",
    destfile="ReactomePathways.gmt.zip")
unzip("ReactomePathways.gmt.zip")
genesets<-gmt_import("ReactomePathways.gmt")

In this cut down example we will be using a sample of 200 Reactome gene sets:

data(genesetsExample)
head(genesetsExample,3)
## $`2-LTR circle formation`
##  [1] "Reactome Pathway" "BANF1"            "HMGA1"            "LIG4"            
##  [5] "PSIP1"            "XRCC4"            "XRCC5"            "XRCC6"           
##  [9] "gag"              "gag-pol"          "rev"              "vif"             
## [13] "vpr"              "vpu"             
## 
## $`5-Phosphoribose 1-diphosphate biosynthesis`
## [1] "Reactome Pathway" "PRPS1"            "PRPS1L1"          "PRPS2"           
## 
## $`A tetrasaccharide linker sequence is required for GAG synthesis`
##  [1] "Reactome Pathway" "AGRN"             "B3GALT6"          "B3GAT1"          
##  [5] "B3GAT2"           "B3GAT3"           "B4GALT7"          "BCAN"            
##  [9] "BGN"              "CSPG4"            "CSPG5"            "DCN"             
## [13] "GPC1"             "GPC2"             "GPC3"             "GPC4"            
## [17] "GPC5"             "GPC6"             "HSPG2"            "NCAN"            
## [21] "SDC1"             "SDC2"             "SDC3"             "SDC4"            
## [25] "VCAN"             "XYLT1"            "XYLT2"

Importing profiling data

mitch accepts pre-ranked data supplied by the user, but also has a function called mitch_import for importing tables generated by limma, edgeR, DESeq2, ABSSeq, Sleuth, Seurat, Muscat and several other upstream tools. By default, only the genes that are detected in all contrasts are included, but this behaviour can be modified for sparse data setting joinType=full. The below example imports two edgeR tables called “rna” and “k9a” Where gene identifiers are present as row names. Note that if there is more than one profile being imported, they need to be part of a list.

data(rna,k9a)
x<-list("rna"=rna,"k9a"=k9a)
y<-mitch_import(x,"edgeR")
## Note: Mean no. genes in input = 1000
## Note: no. genes in output = 1000
## Note: estimated proportion of input genes in output = 1
head(y)
##                  rna        k9a
## NR4A3       68.07237 10.7310010
## HSPA1B      47.19114 18.8135155
## DNAJB1      35.12799  2.4326983
## MIR133A1HG -27.36199  8.9061967
## HSPH1       25.83750 10.8922577
## CXCL2       24.76570  0.8459414

mitch can do unidimensional analysis if you provide it a single profile as a dataframe (not in a list).

y<-mitch_import(rna,DEtype="edger")
## The input is a single dataframe; one contrast only. Converting
##         it to a list for you.
## Note: Mean no. genes in input = 1000
## Note: no. genes in output = 1000
## Note: estimated proportion of input genes in output = 1
head(y)
##                    x
## NR4A3       68.07237
## HSPA1B      47.19114
## DNAJB1      35.12799
## MIR133A1HG -27.36199
## HSPH1       25.83750
## CXCL2       24.76570

If the gene identifiers are not given in the rownames, then the column can be specified with the geneIDcol parameter like this:

# first rearrange cols
rna_mod<-rna
rna_mod$MyGeneIDs<-rownames(rna_mod)
rownames(rna_mod)<-seq(nrow(rna_mod))
head(rna_mod)
##      logFC  logCPM      PValue adj.p.value  MyGeneIDs
## 1  4.12734 5.09552 8.46507e-69 6.24341e-65      NR4A3
## 2  3.64685 7.42834 6.43968e-48 3.16639e-44     HSPA1B
## 3  2.35432 7.13208 7.44748e-36 2.74645e-32     DNAJB1
## 4 -1.02085 7.29935 4.34519e-28 1.28192e-24 MIR133A1HG
## 5  1.11729 6.03741 1.45377e-26 3.06351e-23      HSPH1
## 6  5.48158 2.88719 1.71515e-25 3.16252e-22      CXCL2
# now import with geneIDcol
y<-mitch_import(rna_mod,DEtype="edgeR",geneIDcol="MyGeneIDs")
## The input is a single dataframe; one contrast only. Converting
##         it to a list for you.
## Note: Mean no. genes in input = 1000
## Note: no. genes in output = 1000
## Note: estimated proportion of input genes in output = 1
head(y)
##                    x
## NR4A3       68.07237
## HSPA1B      47.19114
## DNAJB1      35.12799
## MIR133A1HG -27.36199
## HSPH1       25.83750
## CXCL2       24.76570

By default, differential gene activity is scored using a supplied test statistic or directional p-value (D):

D = sgn(logFC) * -log10(p-value)

If this is not desired, then users can perform their own custom scoring procedure and import with DEtype="prescored".

There are many cases where the gene IDs don’t match the gene sets. To overcome this, mitch_import also accepts a two-column table (gt here) that relates gene identifiers in the profiling data to those in the gene sets. In this example we can create some fake gene accession numbers to demonstrate this feature.

library("stringi")
# obtain vector of gene names
genenames<-rownames(rna)
# create fake accession numbers
accessions<-paste("Gene0",stri_rand_strings(nrow(rna)*2, 6, pattern = "[0-9]"),sep="")
accessions<-head(unique(accessions),nrow(rna))
# create a gene table file that relates gene names to accession numbers
gt<-data.frame(genenames,accessions)

# now swap gene names for accessions
rna2<-merge(rna,gt,by.x=0,by.y="genenames")
rownames(rna2)<-rna2$accessions
rna2<-rna2[,2:5]

k9a2<-merge(k9a,gt,by.x=0,by.y="genenames")
rownames(k9a2)<-k9a2$accessions
k9a2<-k9a2[,2:5]

# now have a peek at the input data before importing
head(rna2,3)
##                 logFC  logCPM      PValue adj.p.value
## Gene0868126  0.296028 6.82814 3.84512e-04 1.46941e-02
## Gene0402367 -0.375440 4.71470 1.09120e-03 3.05432e-02
## Gene0478214  0.882624 8.12078 2.11945e-11 8.01642e-09
head(k9a2,3)
##                logFC  logCPM      PValue adj.p.value
## Gene0868126 0.339535 3.67309 1.62925e-03 1.43363e-02
## Gene0402367 0.585837 3.66069 3.23724e-04 4.06552e-03
## Gene0478214 1.138700 2.78713 4.94270e-14 1.69263e-11
head(gt,3)
##   genenames  accessions
## 1     NR4A3 Gene0590582
## 2    HSPA1B Gene0994163
## 3    DNAJB1 Gene0094529
x<-list("rna2"=rna2,"k9a2"=k9a2)
y<-mitch_import(x,DEtype="edgeR",geneTable=gt)
## Note: Mean no. genes in input = 1000
## Note: no. genes in output = 1000
## Note: estimated proportion of input genes in output = 1
head(y,3)
##           rna2      k9a2
## A2M   3.415090  2.788012
## AAAS -2.962096  3.489825
## ABRA 10.673777 13.306036

?mitch_import provides more instructions on using this feature.

Calculating enrichment

The mitch_calc function performs multivariate enrichment analysis of the supplied gene sets in the scored profiling data. At its simpest form mitch_calc function accepts the scored data as the first argument and the genesets as the second argument. Users can prioritise enrichments based on small adjusted p-values, by the observed effect size (magnitude of “s”, the enrichment score) or the standard deviation of the s scores. Note that the number of parallel cores is set here to cores=2 but the default is to use all but one available cores.

# prioritisation by significance
res<-mitch_calc(y,genesetsExample,priority="significance",cores=2)
## Note: When prioritising by significance (ie: small 
##             p-values), large effect sizes might be missed.
# peek at the results
head(res$enrichment_result)
##                                                           set setSize
## 5                                       Biological oxidations      10
## 2 Antigen processing: Ubiquitination & Proteasome degradation      20
## 1                                      Adaptive Immune System      44
## 3                           Asparagine N-linked glycosylation      17
## 4                                               Axon guidance      48
##      pMANOVA       s.rna2      s.k9a2      p.rna2     p.k9a2    s.dist
## 5 0.00899742 -0.555757576 -0.04141414 0.002429897 0.82165390 0.5572985
## 2 0.05970462 -0.055510204  0.28377551 0.670730417 0.02956405 0.2891538
## 1 0.52997174  0.000000000  0.09775580 1.000000000 0.27259775 0.0977558
## 3 0.53769811 -0.136137873  0.04589791 0.335578518 0.74549830 0.1436668
## 4 0.94522077  0.006959034 -0.02551646 0.935141485 0.76540436 0.0264484
##           SD p.adjustMANOVA
## 5 0.36369573      0.0449871
## 2 0.23991123      0.1492616
## 1 0.06912379      0.6721226
## 3 0.12871874      0.6721226
## 4 0.02296364      0.9452208
# prioritisation by effect size
res<-mitch_calc(y,genesetsExample,priority="effect",cores=2)
## Note: Enrichments with large effect sizes may not be 
##             statistically significant.
head(res$enrichment_result)
##                                                           set setSize
## 5                                       Biological oxidations      10
## 2 Antigen processing: Ubiquitination & Proteasome degradation      20
## 3                           Asparagine N-linked glycosylation      17
## 1                                      Adaptive Immune System      44
## 4                                               Axon guidance      48
##      pMANOVA       s.rna2      s.k9a2      p.rna2     p.k9a2    s.dist
## 5 0.00899742 -0.555757576 -0.04141414 0.002429897 0.82165390 0.5572985
## 2 0.05970462 -0.055510204  0.28377551 0.670730417 0.02956405 0.2891538
## 3 0.53769811 -0.136137873  0.04589791 0.335578518 0.74549830 0.1436668
## 1 0.52997174  0.000000000  0.09775580 1.000000000 0.27259775 0.0977558
## 4 0.94522077  0.006959034 -0.02551646 0.935141485 0.76540436 0.0264484
##           SD p.adjustMANOVA
## 5 0.36369573      0.0449871
## 2 0.23991123      0.1492616
## 3 0.12871874      0.6721226
## 1 0.06912379      0.6721226
## 4 0.02296364      0.9452208

By default, gene sets with fewer than 10 members present in the profiling data are discarded. This threshold can be modified using the minsetsize option. There is no upper limit of gene set size.

res<-mitch_calc(y,genesetsExample,priority="significance",minsetsize=5,cores=2)
## Note: When prioritising by significance (ie: small 
##             p-values), large effect sizes might be missed.

By default, in downstream visualisation steps, charts are made from the top 50 gene sets, but this can be modified using the resrows option.

res<-mitch_calc(y,genesetsExample,priority="significance",resrows=3,cores=2)
## Note: When prioritising by significance (ie: small 
##             p-values), large effect sizes might be missed.

Generate a HTML report

The HTML reports contain several plots as raster images and interactive charts which are useful as a first-pass visualisation. These can be generated like this:

mitch_report(res,"myreport.html")
## Dataset saved as " /tmp/RtmpXv1Dxf/myreport.rds ".
## processing file: mitch.Rmd
## output file: /tmp/Rtmpm4Ddcs/Rbuild28d711a88fa9/mitch/vignettes/mitch.knit.md
## 
## Output created: /tmp/RtmpXv1Dxf/mitch_report.html

Generate high resolution plots

In case you want the charts in PDF format, for example for publications, these can be generated as such:

mitch_plots(res,outfile="mycharts.pdf")

Session Info

sessionInfo()
## R version 4.0.2 (2020-06-22)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 18.04.4 LTS
## 
## Matrix products: default
## BLAS:   /home/biocbuild/bbs-3.11-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.11-bioc/R/lib/libRlapack.so
## 
## locale:
##  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=C              
##  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
##  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] pkgload_1.1.0   GGally_2.0.0    ggplot2_3.3.2   reshape2_1.4.4 
##  [5] beeswarm_0.2.3  gplots_3.1.0    gtools_3.8.2    tibble_3.0.3   
##  [9] dplyr_1.0.2     echarts4r_0.3.2 stringi_1.5.3   mitch_1.0.10   
## 
## loaded via a namespace (and not attached):
##  [1] Rcpp_1.0.5         assertthat_0.2.1   rprojroot_1.3-2    digest_0.6.25     
##  [5] mime_0.9           R6_2.4.1           plyr_1.8.6         backports_1.1.10  
##  [9] evaluate_0.14      highr_0.8          pillar_1.4.6       rlang_0.4.8       
## [13] rmarkdown_2.4      desc_1.2.0         labeling_0.3       stringr_1.4.0     
## [17] htmlwidgets_1.5.2  munsell_0.5.0      shiny_1.5.0        compiler_4.0.2    
## [21] httpuv_1.5.4       xfun_0.18          pkgconfig_2.0.3    htmltools_0.5.0   
## [25] tidyselect_1.1.0   gridExtra_2.3      reshape_0.8.8      crayon_1.3.4      
## [29] withr_2.3.0        later_1.1.0.1      MASS_7.3-53        bitops_1.0-6      
## [33] grid_4.0.2         jsonlite_1.7.1     xtable_1.8-4       gtable_0.3.0      
## [37] lifecycle_0.2.0    magrittr_1.5       scales_1.1.1       KernSmooth_2.23-17
## [41] farver_2.0.3       promises_1.1.1     testthat_2.3.2     ellipsis_0.3.1    
## [45] generics_0.0.2     vctrs_0.3.4        RColorBrewer_1.1-2 tools_4.0.2       
## [49] glue_1.4.2         purrr_0.3.4        parallel_4.0.2     fastmap_1.0.1     
## [53] yaml_2.2.1         colorspace_1.4-1   caTools_1.18.0     knitr_1.30