OPPAR: Outlier Profile and Pathway Analysis in R

Soroor Hediyehzadeh

2017-10-30

Cancer Outlier Profile Analysis (COPA) is a common analysis to identify genes that might be down-regulated or up-regulated only in a proportion of samples with the codition of interest. OPPAR is the R implementation of modified COPA (mCOPA) method, originally published by Chenwei Wang et al. in 2012. The aim is to identify genes that are outliers in samples with condition of interest, compared to normal samples. The methods implemented in OPPAR enable the users to perform the analysis in various ways, namely detecting outlier features in control versus condition samples (whether or not there is a information on subtypes), and detecting genes that are outlier in one subtype compared to the other samples, if the subtypes are known.

OPPAR can also be used for Gene Set Enrichment Analysis (GSEA). Here, a modified version of GSVA method is implemented. GSVA can be used to determine which samples in the study are enriched for gene expression signatures that are of interest. The gsva function in GSVA package returns an enrichment score for each sample, for the given signatures/gene sets. With the current implementation of the method, samples that strongly show enrichment for down(-regulated) gene expression signatures will receive negative scores. However, Often it is in the interest of the biologists and researchers to get positive scores for samples that are enriched in both up and down signatures. Therefore, the gsva function has been modified to assign positive scores to samples that are enriched for the up-regulated and down-regulated gene expression signatures.

OPPAR comes with four functions:

This vignette illustrates a possible workflow for OPPAR, using Tomlins et al. prostate cancer data. In addition, Maupin’s TGFb data have been analyzed for enrichement of a TGFb gene signature in the samples measured in this study.

Please note although the analysis presented here have been done on microarray studies, one can apply oppar tools to RPKM values of gene expression measurements in NGS studies.


Analysis of Tomlins et al. prostate cancer dataset

Data was retrieved from GEO database, checked for normalization and subsetted according to procedure outlined in the mCOPA paper. In addition, probes with no annotation were removed. The impute package was used to impute the missing values using K-nearest neighbours method (k = 10). The subsetted dataset is available in the package as a sample data, and contains an ExpressionSet object, storing information on samples, genes and gene expressions. We apply opa on Tomlins et al. data, then use getSubtypeProbes to get all down- regulated and all up-regulated outliers.

opa returns the outlier profile matrix, which is a matrix of -1 ( for down-regulated outliers), 0 ( not an outlier) and 1 (up-regulated outlier). For more information see ?opa

For a brief overview of oppar package and functions, please see ?oppar

## Loading required package: knitr
data(Tomlins) # loads processed Tomlins data
## Warning in data(Tomlins): data set 'Tomlins' not found
# the first 21 samples are Normal samples, and the rest of 
# the samples are our cases (metastatic). We, thus, generate a group
# variable for the samples based on this knowledge.

g <- factor(c(rep(0,21),rep(1,ncol(exprs(eset)) - 21)))
g
##  [1] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1
## [36] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
## [71] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
## Levels: 0 1
# Apply opa on Tomlins data, to detect outliers relative to the
# lower 10% (lower.quantile = 0.1) and upper 5% (upper.quantile = 0.95 -- Default) of 
# gene expressions.
tomlins.opa <- opa(eset, group = g, lower.quantile = 0.1)
tomlins.opa
## Object of type OPPARList
## Features: 663
## Samples: 65
## Upper quantile: 0.95
## Lower quantile: 0.10
## Groups:
##  [1] 1 1 1 1 1 1 1 1 1 1
## Levels: 0 1

The matrix containing the outlier profiles is called profileMatrix and can be accessed using the $ operator. The upper.quantile and lower.quantile parameters used to run the function can also be retrieved using this operator.

tomlins.opa$profileMatrix[1:6,1:5]
##             GSM141341 GSM141342 GSM141343 GSM141356 GSM141357
## Hs6-1-10-7          0         0         0         0         0
## Hs6-1-11-3          0         0         0         0         0
## Hs6-1-12-24         0         0         0         0         0
## Hs6-1-12-8          0         0         0         0         0
## Hs6-1-13-12         0         0         0         0         0
## Hs6-1-13-13         0         0         0         0         0
tomlins.opa$upper.quantile
## [1] 0.95
tomlins.opa$lower.quantile
## [1] 0.1

We can extract outlier profiles for any individual samples in the profileMatrix, using getSampleOutlier. see ?getSampleOutlier for more detailed information

getSampleOutlier(tomlins.opa, c(1,5))
## $GSM141341.up
## [1] "Hs6-26-18-3"
## 
## $GSM141341.down
## [1] "Hs6-17-9-11" "Hs6-2-11-4" 
## 
## $GSM141357.up
##  [1] "Hs6-10-22-5"  "Hs6-14-25-7"  "Hs6-15-22-3"  "Hs6-15-5-11" 
##  [5] "Hs6-16-2-15"  "Hs6-16-9-2"   "Hs6-18-9-13"  "Hs6-20-13-20"
##  [9] "Hs6-20-7-7"   "Hs6-22-21-5"  "Hs6-26-23-5"  "Hs6-28-5-5"  
## 
## $GSM141357.down
## [1] "Hs6-13-3-14"  "Hs6-22-5-9"   "Hs6-24-7-23"  "Hs6-25-12-19"
## [5] "Hs6-25-2-9"

Extracting down-regulated and up-regulated outliers in all samples using getSubtypeProbes:

outlier.list <- getSubtypeProbes(tomlins.opa, 1:ncol(tomlins.opa$profileMatrix))

We can then obtain a list of GO terms from org.Hs.eg.db. Each element of the list will be a GO terms with the Entrez gene IDs corresponding to that term. We can the apply mroast from limma package for multiple gene set enrichment testing.

# gene set testing with limma::mroast
#biocLite(org.Hs.eg.db)
library(org.Hs.eg.db)
library(limma)
org.Hs.egGO2EG
## GO2EG map for Human (object of class "Go3AnnDbBimap")
go2eg <- as.list(org.Hs.egGO2EG)
head(go2eg)
## $`GO:0000002`
##     TAS     IMP     ISS     IMP     IMP     NAS     IMP     IDA     IBA 
##   "291"  "1890"  "4205"  "4358"  "4976"  "9361" "10000" "55186" "63875" 
##     IBA     IDA     IMP 
## "80119" "84275" "92667" 
## 
## $`GO:0000003`
##    IEP 
## "8510" 
## 
## $`GO:0000011`
##     IBA 
## "64145" 
## 
## $`GO:0000012`
##         IDA         IDA         IEA         IMP         IDA         IDA 
##      "3981"      "7141"      "7515"     "23411"     "54840"     "55775" 
##         IMP         IMP         IEA 
##     "55775"    "200558" "100133315" 
## 
## $`GO:0000018`
##     TAS     TAS     TAS     IMP     IMP     IEP 
##  "3575"  "3836"  "3838"  "9984" "10189" "56916" 
## 
## $`GO:0000019`
##     TAS     IDA 
##  "4361" "10111"
# Gene Set analysis using rost from limma

# need to subset gene express data based on up outliers
up.mtrx <- exprs(eset)[fData(eset)$ID %in% outlier.list[["up"]], ]
# get Entrez gene IDs for genes in up.mtrx

entrez.ids.up.mtrx <- fData(eset)$Gene.ID[fData(eset)$ID %in% rownames(up.mtrx)]

# find the index of genes in GO gene set in the gene expression matrix
gset.idx <- lapply(go2eg, function(x){
    match(x, entrez.ids.up.mtrx)
})

# remove missing values
gset.idx <- lapply(gset.idx, function(x){
    x[!is.na(x)]
})

# removing gene sets with less than 10 elements
gset.ls <- unlist(lapply(gset.idx, length))
gset.idx <- gset.idx[which(gset.ls > 10)]

# need to define a model.matrix for mroast
design <- model.matrix(~ g)
up.mroast <- mroast(up.mtrx, index = gset.idx, design = design) 
head(up.mroast, n=5)
##            NGenes PropDown    PropUp Direction PValue   FDR PValue.Mixed
## GO:0005576     15        0 1.0000000        Up  0.001 0.001        0.001
## GO:0051301     13        0 1.0000000        Up  0.001 0.001        0.001
## GO:0005615     13        0 1.0000000        Up  0.001 0.001        0.001
## GO:0007165     20        0 0.9500000        Up  0.001 0.001        0.001
## GO:0042803     12        0 0.9166667        Up  0.001 0.001        0.001
##            FDR.Mixed
## GO:0005576     0.001
## GO:0051301     0.001
## GO:0005615     0.001
## GO:0007165     0.001
## GO:0042803     0.001

The GO terms for the first 10 GO Ids detected by mroast can be retrieved in the following way.

go.terms <- rownames(up.mroast[1:10,])
#biocLite(GO.db)
library(GO.db)
columns(GO.db)
## [1] "DEFINITION" "GOID"       "ONTOLOGY"   "TERM"
keytypes(GO.db)
## [1] "DEFINITION" "GOID"       "ONTOLOGY"   "TERM"
r2tab <- select(GO.db, keys=go.terms,
                columns=c("GOID","TERM"), 
                keytype="GOID")
r2tab
##          GOID                              TERM
## 1  GO:0005576              extracellular region
## 2  GO:0051301                     cell division
## 3  GO:0005615               extracellular space
## 4  GO:0007165               signal transduction
## 5  GO:0042803 protein homodimerization activity
## 6  GO:0016021    integral component of membrane
## 7  GO:0003723                       RNA binding
## 8  GO:0005515                   protein binding
## 9  GO:0005737                         cytoplasm
## 10 GO:0005524                       ATP binding

We repeating the above steps for down-regulated outliers, to see what GO terms they are enriched for.

library(org.Hs.eg.db)
library(limma)
org.Hs.egGO2EG
## GO2EG map for Human (object of class "Go3AnnDbBimap")
go2eg <- as.list(org.Hs.egGO2EG)
head(go2eg)
## $`GO:0000002`
##     TAS     IMP     ISS     IMP     IMP     NAS     IMP     IDA     IBA 
##   "291"  "1890"  "4205"  "4358"  "4976"  "9361" "10000" "55186" "63875" 
##     IBA     IDA     IMP 
## "80119" "84275" "92667" 
## 
## $`GO:0000003`
##    IEP 
## "8510" 
## 
## $`GO:0000011`
##     IBA 
## "64145" 
## 
## $`GO:0000012`
##         IDA         IDA         IEA         IMP         IDA         IDA 
##      "3981"      "7141"      "7515"     "23411"     "54840"     "55775" 
##         IMP         IMP         IEA 
##     "55775"    "200558" "100133315" 
## 
## $`GO:0000018`
##     TAS     TAS     TAS     IMP     IMP     IEP 
##  "3575"  "3836"  "3838"  "9984" "10189" "56916" 
## 
## $`GO:0000019`
##     TAS     IDA 
##  "4361" "10111"
# subsetting gene expression matrix based on down outliers
down_mtrx <- exprs(eset)[fData(eset)$ID %in% outlier.list[["down"]], ]
entrez_ids_down_mtrx <- fData(eset)$Gene.ID[fData(eset)$ID %in% rownames(down_mtrx)]

gset_idx_down <- lapply(go2eg, function(x){
    match(x, entrez_ids_down_mtrx)
})

# remove missing values
gset_idx_down <- lapply(gset_idx_down, function(x){
    x[!is.na(x)]
})

# removing gene sets with less than 10 elements
gset_ls_down <- unlist(lapply(gset_idx_down, length))
gset_idx_down <- gset_idx_down[which(gset_ls_down > 10)]

# apply mroast
down_mroast <- mroast(down_mtrx, gset_idx_down, design) 
head(down_mroast, n=5)
##            NGenes PropDown PropUp Direction PValue   FDR PValue.Mixed
## GO:0005615     14        1      0      Down  0.001 0.001        0.001
## GO:0046872     14        1      0      Down  0.001 0.001        0.001
## GO:0008270     13        1      0      Down  0.001 0.001        0.001
## GO:0007165     12        1      0      Down  0.001 0.001        0.001
## GO:0005739     11        1      0      Down  0.001 0.001        0.001
##            FDR.Mixed
## GO:0005615     0.001
## GO:0046872     0.001
## GO:0008270     0.001
## GO:0007165     0.001
## GO:0005739     0.001

And extract GO terms for the top 10 results:

go_terms_down <- rownames(down_mroast[1:10,])

dr2tab <- select(GO.db, keys=go_terms_down,
                columns=c("GOID","TERM"), 
                keytype="GOID")
dr2tab
##          GOID                                  TERM
## 1  GO:0005615                   extracellular space
## 2  GO:0046872                     metal ion binding
## 3  GO:0008270                      zinc ion binding
## 4  GO:0007165                   signal transduction
## 5  GO:0005739                         mitochondrion
## 6  GO:0005794                       Golgi apparatus
## 7  GO:0005887 integral component of plasma membrane
## 8  GO:0005829                               cytosol
## 9  GO:0005886                       plasma membrane
## 10 GO:0016020                              membrane

Gene Set Enrichment Analysis

We are now going to perform enrichment analysis for on Maupin’s TGFb data (see ?maupin), given a gene signature. The maupin data object contains a matrix containing gene expression measurements on 3 control samples and 3 TGFb induced samples. We run the modified gsva function introduced in this package to get one large positive scores for samples enriched in the given gene signature, both for down gene signature and up gene signature. This is while the original gsva function returns negative scores for samples that are enriched in down gene signature, and positive scores for samples enriched in up gene signature. Therefore, the scores returned by the gsva function in this package are the sum of the scores for up gene signature and down gene signature. Note that in order for the modified version of the gsva function to work properly, the gset.idx.list has to be a named list, with the up signature gene list being named ‘up’ and down gene signature gene list being names ‘down’ (see example code below). Also note that the is.gset.list.up.down argument has to be set to TRUE if the user wishes to use the modified version (i.e to get the sum of es scores for up and down gene signatures). See ?gsva for more details.

data("Maupin")
names(maupin)
## [1] "data" "sig"
head(maupin$data)
##    M_Ctrl_R1 M_Ctrl_R2 M_Ctrl_R3 M_TGFb_R1 M_TGFb_R2 M_TGFb_R3
## 2   4.551955  4.391799  4.306602  4.738577  4.579810  4.576038
## 9   7.312850  7.155411  7.274249  7.520725  7.381180  7.279445
## 10  4.699286  4.625667  4.624420  4.613213  4.779147  4.845243
## 12  5.552299  5.806786  5.891174  6.976169  7.169206  7.200424
## 13  5.524958  5.341497  5.422172  5.569487  5.597191  5.585326
## 14  9.025984  9.075223  8.951924  9.062231  9.130251  9.204617
head(maupin$sig)
##   EntrezID  Symbol upDown_integrative_signature
## 1       19   ABCA1                            0
## 2       87   ACTN1                            0
## 3      136 ADORA2B                            0
## 4      182    JAG1                           up
## 5      220 ALDH1A3                         down
## 6      224 ALDH3A2                            0
##   upDown_comparative_signature upDown
## 1                           up     up
## 2                           up     up
## 3                         down   down
## 4                           up     up
## 5                            0   down
## 6                         down   down
geneSet<- maupin$sig$EntrezID    #Symbol  ##EntrezID # both up and down genes:

up_sig<- maupin$sig[maupin$sig$upDown == "up",]
d_sig<- maupin$sig[maupin$sig$upDown == "down",]
u_geneSet<- up_sig$EntrezID   #Symbol   # up_sig$Symbol  ## EntrezID
d_geneSet<- d_sig$EntrezID

enrichment_scores <- gsva(maupin$data, list(up = u_geneSet, down= d_geneSet), mx.diff=1,
               verbose=TRUE, abs.ranking=FALSE, is.gset.list.up.down=TRUE, parallel.sz = 1 )$es.obs
## Estimating GSVA scores for 2 gene sets.
## Computing observed enrichment scores
## Estimating ECDFs in microarray data with Gaussian kernels
## Using parallel with 1 cores
## 
  |                                                                       
  |                                                                 |   0%
  |                                                                       
  |================================                                 |  50%
  |                                                                       
  |=================================================================| 100%
head(enrichment_scores)
##          M_Ctrl_R1  M_Ctrl_R2  M_Ctrl_R3 M_TGFb_R1 M_TGFb_R2 M_TGFb_R3
## GeneSet -0.8991905 -0.7841492 -0.8329552 0.9041564 0.7714735 0.8147947
## calculating enrichment scores using ssgsea method
# es.dif.ssg <- gsva(maupin, list(up = u_geneSet, down= d_geneSet),
#                                                        verbose=TRUE, abs.ranking=FALSE, is.gset.list.up.down=TRUE,
#                                                        method = "ssgsea")

A histogram of enrichment scores is plotted below and the density of es scores for TGFb samples is shown in red. The distribution of es scores of control samples is shown in blue. As it can be seen from the plot below, TGFb induced samples that are expected to be enriched in the given TGFb signature have received positive scores and are on the right side of the histogram, whereas the control samples are on the left side of the histogram. In addition, TGFb induced samples and contorl samples have been nicely separated from each other.

hist(enrichment_scores, main = "enrichment scores", xlab="es")
lines(density(enrichment_scores[,1:3]), col = "blue") # control samples
lines(density(enrichment_scores[,4:6]), col = "red") # TGFb samples
legend("topleft", c("Control","TGFb"), lty = 1, col=c("blue","red"), cex = 0.6)