knitr::opts_chunk$set( collapse = TRUE, cache = FALSE, comment = "#>" )
library(knitr) library(papeR) figDir_char <- system.file(package = "Bioc2019pathwayPCA", "vignette")
With the advance in high-throughput technology for molecular assays, multi-omics datasets have become increasingly available. In this workshop, we will demonstrate using the pathwayPCA package to perform integrative pathway-based analyses of multi-omics datasets. In particular, we will demonstrate through three case studies the capabilities of pathwayPCA
to
There is a built version of this workshop available at RPubs.^[http://rpubs.com/gabrielodom/Bioc2019_pathwayPCA] The source is available on GitHub.^[https://github.com/gabrielodom/Bioc2019pathwayPCA]
This will be a hands-on workshop, students will live-code with us. It would be helpful for participants to bring a laptop with RStudio and pathwayPCA package installed.^[https://gabrielodom.github.io/pathwayPCA/index.html#installing-the-package]
To install the workshop dependencies, assuming you have already installed Bioconductor:^[https://www.bioconductor.org/install]
install.packages(c("survival", "survminer", "tidyverse")) BiocManager::install("rWikiPathways") BiocManager::install("pathwayPCA")
We also strongly recommend the RStudio^[https://www.rstudio.com/products/rstudio/download/] integrated development environment as a user-friendly graphical wrapper for R. We require R version 3.6 or later.
This workshop package, Bioc2019pathwayPCA
, depends on two packages:
pathwayPCA
,^[https://gabrielodom.github.io/pathwayPCA/] and tidyverse
.^[https://www.tidyverse.org/]The tidyverse
package suite is a collection of utility packages for data science: ggplot2
, dplyr
, tidyr
, readr
, purrr
, tibble
, stringr
, and forcats
. We will make use of some data constructs and ideas from these packages, but we do not expect users to be intimately familiar with them.
Additionally, because we will use survival data in our examples, we need survival
^[https://CRAN.R-project.org/package=survival] and survminer
^[https://cran.r-project.org/package=survminer] (for drawing survival curves with ggplot2
).
library(survival) library(survminer) library(Bioc2019pathwayPCA)
Because we are currently in the development phase for version 2 of this package, you can install the package from GitHub. In order to install a package from GitHub, you will need the devtools::
^[https://github.com/r-lib/devtools] package and either Rtools^[https://cran.r-project.org/bin/windows/Rtools/] (for Windows) or Xcode^[https://developer.apple.com/xcode/] (for Mac). Then you can install the development version of the pathwayPCA
package from GitHub via:
devtools::install_github("gabrielodom/pathwayPCA")
50m total
| Activity | Time |
|----------------------------------------------------------------------|------|
| Introduction to pathwayPCA
| 5m |
| Case Study 1: Estimating sample-specific pathway activities | 15m |
| Case Study 2: Pathway analysis for multi-omics datasets | 10m |
| Case Study 3: Analyzing experiments with covariates and interactions | 10m |
| Summary and Conclusion | 5m |
| Questions and Comments | 5m |
pathwayPCA
workflow to typical gene expression and copy number variations datapathwayPCA
to experiments with complex designs, such as those with covariate and/or interaction effectspathwayPCA
approachpathwayPCA
pathwayPCA
is an integrative analysis tool that implements PCA-based pathway analysis approaches.^[ Chen et al. (2008) ] ^[ Chen et al. (2010) ] ^[ Chen (2011) ]
pathwayPCA
pathwayPCA
allows users to:
How does pathwayPCA
fit into the plethora of pathway analysis tools available?
pathwayPCA
tests the self-contained null hypothesis (Q1) (PMID: 21565265^[https://doi.org/10.1016/j.ygeno.2011.04.006]); that is, the features (e.g. genes) in a pathway are not associated with the disease phenotype.goseq
,^[https://bioconductor.org/packages/release/bioc/html/goseq.html] perform over-representation analysis and test for competitive null hypothesis (Q2); that is, the genes in a pathway show the same magnitude of associations with the disease phenotype compared with genes in the rest of the genome.pathwayPCA
models correlations within pathways when constructing PCA scores. globaltest
,^[https://doi.org/10.18129/B9.bioc.globaltest] ignore expert-based biological information and fail to consider gene-gene correlations contained in biological pathways.MOFA
,^[https://bioconductor.org/packages/devel/bioc/html/MOFA.html] mixOmics
,^[https://www.bioconductor.org/packages/release/bioc/html/mixOmics.html] mogsa
^[https://bioconductor.org/packages/release/bioc/html/mogsa.html], also perform integrative analysis using PCA. However, there is one important difference: these tools typically extract PCs on all the genes in dataset, and then map selected genes (e.g. those with nonzero loadings) to each pathway. In contrast, pathwayPCA
subsets genes for each pathway first, then extracts PCs for each pathway: this helps to reduce noisy signals from irrelevant genes and improves sensitivity and specificity (on association with phenotype) of the PCs.pathwayPCA
provides seamless multi-omics integration via estimated subject-specific pathway activityCoGAPS
^[https://www.bioconductor.org/packages/release/bioc/html/CoGAPS.html] (requires OpenMP) or MOFA
(requires Python).Caveats for using pathwayPCA
:
pathwayPCA
is a poor choice for GISTIC calls (copy number data) or mutation data.include_graphics(file.path(figDir_char, "pathwayPCA_workflow.png"))
pathwayPCA
uses four main groups of functions: data import and wrangling, object creation, object analysis, and analysis inspection. The main function groups are
read_gmt()
imports a .gmt
file as a pathway collection.SE2Tidy()
extracts an assay from a SummarizedExperiment
object,^[https://doi.org/10.18129/B9.bioc.SummarizedExperiment] and turns it into a "tidy" data frame.TransposeAssay()
is a variant of the base t()
function designed specifcially for data frames and tibbles. It preserves row and column names after transposition.Omics*
Objects:CreateOmics()
takes in a collection of pathways, a single -omics assay, and a clinical response data frame and creates a data object of class Omics*
.SubsetPathwayData()
can extract the pathway-specific assay values and responses for a given pathway from an Omics*
object.AESPCA_pVals()
takes in an Omics*
object and calculates pathway $p$-values (parametrically or non-parametrically), principal components, and loadings via AESPCA. This returns an object of class aespcOut
.SuperPCA_pVals()
takes in an Omics*
object with valid response information and calculates pathway parametric $p$-values, principal components, and loadings via SuperPCA. This returns an object of class superpcOut
.getPathPCLs()
takes in an object of class aespcOut
or superpcOut
and the TERMS
name of a pathway. This function extracts 1) the data frame of principal components and subject IDs for the given pathway, and 2) a data frame of sparse loadings and feature names for the given pathway.getPathpVals()
takes in an object of class aespcOut
or superpcOut
and returns a table of the $p$-values and false discovery rates for each pathway.pathwayPCA
inputsWe define our assay data format in R as follows:
Use the two utility functions, TransposeAssay()
or SE2Tidy()
, to wrangle your assay data into appropriate form for further analysis. Here is an example of "tidying" the first assay from a SummarizedExperiment
object:
# DONT RUN library(SummarizedExperiment) data(airway, package = "airway") airway_df <- SE2Tidy(airway, whichAssay = 1)
Based on expert knowledge of the problem at hand, curate a set of biological pathways $1, \ldots, K$, where pathway $k$ contains $p_k$ -ome measurements. These pathway collections are often stored in a .gmt
file, a text file with each row corresponding to one pathway. Each row contains an ID (column TERMS
), an optional description (column description
), and the genes in the pathway (all subsequent columns). This file format is independently defined by the Broad Institute.^[http://software.broadinstitute.org/cancer/software/gsea/wiki/index.php/Data_formats] Pathway collections in .gmt
form can be downloaded from the MSigDB database.^[http://software.broadinstitute.org/gsea/msigdb/collections.jsp]
For WikiPathways,^[https://www.wikipathways.org/index.php/WikiPathways] a collection of community-defined biological pathways, one can download monthly data releases in .gmt
format using the dowloadPathwayArchive()
function in the rWikiPathways
package from Bioconductor.^[https://bioconductor.org/packages/release/bioc/html/rWikiPathways.html] For example, the following commands^[https://bioconductor.org/packages/release/bioc/vignettes/rWikiPathways/inst/doc/Overview.html#4_give_me_more] download the latest release of the human pathways from WikiPathways to your current directory (we do not execute this code):
# DONT RUN library(rWikiPathways) # library(XML) # necessary if you encounter an error with readHTMLTable downloadPathwayArchive( organism = "Homo sapiens", format = "gmt" )
trying URL 'http://data.wikipathways.org/current/gmt/wikipathways-20190110-gmt-Homo_sapiens.gmt' Content type '' length 174868 bytes (170 KB) downloaded 170 KB
print("wikipathways-20190110-gmt-Homo_sapiens.gmt")
pathwayPCA
includes the June 2018 Wikipathways collection for homo sapiens, which can be loaded using the read_gmt
function:
dataDir_path <- system.file( "extdata", package = "pathwayPCA", mustWork = TRUE ) wikipathways_PC <- read_gmt( paste0(dataDir_path, "/wikipathways_human_symbol.gmt"), description = TRUE )
This pathway collection contains:
pathways
: a list of character vectors matching some of the names of the -omes recorded in the assayTERMS
: a character vector of the names of the pathwaysdescription
: (OPTIONAL) descriptions or URLs of the pathwayswikipathways_PC
Pathway-based Adaptive, Elastic-net, Sparse PCA (AESPCA)^[ Chen (2011) ] combines the following methods into a single objective function: Elastic-Net^[ Zou and Hastie, 2005 ], Adaptive Lasso^[ Zou, 2012 ], and Sparse PCA^[ Zou, Hastie, and Tibshirani, 2012 ]. AESPCA extracts principal components from pathway $k$ which minimize this composite objective function. These extracted PCs represent activities within each pathway. The estimated latent variables are then tested against phenotypes using either a parametric or permutation test (permuting the sample responses). Note that the AESPCA approach does not use the response information to estimate pathway PCs, so it is an unsupervised approach.
Pathway-based Supervised PCA (SuperPCA):^[ Chen et al. (2008) ] ^[ Chen et al. (2010) ] ranks each feature in pathway $k$ by its univariate relationship with the outcome of interest (survival time, tumor size, cancer subtype, etc.), then extracts principal components from the features most related to the outcome. Because of this gene selection step, this method is "supervised". Therefore, test statistics from the SuperPCA model can no longer be approximated using the Student's $t$-distribution. To account for the gene selection step, SuperPCA as implemented in pathwayPCA
estimates $p$-values from a two-component mixture of Gumbel extreme value distributions instead.
For this example, we will use the mass spectrometry based global proteomics data for ovarian cancer recently generated by the Clinical Proteomic Tumor Analysis Consortium (CPTAC).^[https://proteomics.cancer.gov/programs/cptac] The gene-level, log-ratio normalized protein abundance expression dataset for tumor samples can be obtained from the LinkedOmics database.^[http://linkedomics.org/data_download/TCGA-OV/] We used the dataset “Proteome (PNNL, Gene level)” which was generated by the Pacific Northwest National Laboratory (PNNL).^[https://www.pnnl.gov/] One subject was removed due to missing survival outcome. Proteins with greater than 20% missingness were removed from the assay. The remaining missing values were imputed using the Bioconductor package impute
under default settings. The final dataset consisted of 5162 protein expression values for 83 samples.
Omics
data object for pathway analysisFirst, we need to create an Omics
-class data object that stores
(1) the expression dataset (2) phenotype information for the samples (3) a collection of pathways
We can obtain protein expression and phenotype datasets for the TCGA ovarian cancer dataset by downloading the raw files from the LinkedOmics website. However, for ease and to save time, we include cleaned and imputed versions of these datasets within this package (for the cleaning process, please see the clean_multi_omics.R
file in inst/scripts
).
The ovProteome_df
dataset is a data frame with protein expression levels and TCGA sample IDs. The variables (columns) include expression data for 4763 proteins for each of the 83 primary tumour samples.
data("ovProteome_df") ovProteome_df[, 1:5]
The ovPheno_df
data frame contains TCGA sample IDs, overall survival time, and overall survival censoring status.
data("ovPheno_df") ovPheno_df
OmicsSurv
data containerNow that we have these three data components (pathway collection, proteomics, and clinical responses), we create an OmicsSurv
data container. Note that when assayData_df
and response
are supplied from two different files, the user must match and merge these data sets by sample IDs. The assay and response must have matching row names, otherwise the function will error.
ov_OmicsSurv <- CreateOmics( # protein expression data assayData_df = ovProteome_df, # pathway collection pathwayCollection_ls = wikipathways_PC, # survival phenotypes response = ovPheno_df, # phenotype is survival data respType = "survival", # retain pathways with > 5 proteins minPathSize = 5 )
To see a summary of the Omics
data object we just created, simply type the name of the object:
ov_OmicsSurv
Once we have a valid Omics
object, we can perform pathway analysis using the AESPCA or SuperPCA methodology as described above. Because the syntax for performing SuperPCA is nearly identical to the AESPCA syntax, we will illustrate only the AESPCA workflow below.
We estimate pathway significance via the following model: phenotype ~ intercept + PC1
. Pathway $p$-values are estimated based on a likelihood ratio test that compares this model to the null model phenotype ~ intercept
. Note that when the value supplied to the numReps
argument is greater than 0, the AESPCA_pvals()
function employs a non-parametric, permutation-based test to assign the statistical significance of this model.
ovarian_aespcOut <- AESPCA_pVals( # The Omics data container object = ov_OmicsSurv, # One principal component per pathway numPCs = 1, # Use parallel computing with 2 cores parallel = TRUE, numCores = 2, # # Use serial computing # parallel = FALSE, # Estimate the p-values parametrically (AESPCA only) numReps = 0, # Control FDR via Benjamini-Hochberg adjustment = "BH" )
This ovarian_aespcOut
object contains 3 components: a table of pathway $p$-values, AESPCA-estimated PCs of each sample from each pathway, and the loadings of each protein onto the AESPCs.
names(ovarian_aespcOut)
We haven't yet added a print()
method for these analysis outputs, so be careful: use the two accessor functions, getPathpVals()
and getPathPCLs()
instead. We will show examples of these functions next.
For this ovarian cancer dataset, the top 20 most significant pathways identified by AESPCA are:
getPathpVals(ovarian_aespcOut)
kable( getPathpVals(ovarian_aespcOut) )
Before constructing a graph of the $p$-values, we extract the top 10 pathways (the default value for numPaths
is 20). The score = TRUE
returns the negative natural logarithm of the unadjusted $p$-values for each pathway (to enhance graphical display of the top pathways).
ovOutGather_df <- getPathpVals(ovarian_aespcOut, score = TRUE, numPaths = 10)
Now we plot the pathway significance level for the top 20 pathways.
ggplot(ovOutGather_df) + # set overall appearance of the plot theme_bw() + # Define the dependent and independent variables aes(x = reorder(terms, score), y = score) + # From the defined variables, create a vertical bar chart geom_col(position = "dodge", fill = "#005030", width = 0.7) + # Add a line showing the alpha = 0.0001 level geom_hline(yintercept = -log10(0.0001), size = 1, color = "#f47321") + # Add pathway labels geom_text( aes(x = reorder(terms, score), label = reorder(description, score), y = 0.1), color = "white", size = 4, hjust = 0 ) + # Set main and axis titles ggtitle("AESPCA Significant Pathways: Ovarian Cancer") + xlab("Pathways") + ylab("Negative Log10 (p-Value)") + # Flip the x and y axes coord_flip()
Because pathways are defined a priori, typically only a subset of genes within each pathway are relevant to the phenotype and contribute to pathway significance. In AESPCA, these relevant genes are the genes with nonzero loadings in the first PC of AESPCs.
For example, we know that the IL-1 Signaling Pathway is one of cancer's "usual suspects".^[https://doi.org/10.1371/journal.pcbi.1003470] For the "IL-1 signaling pathway" (Wikipathways WP195^[https://www.wikipathways.org/index.php/Pathway:WP195]), we can extract the PCs and their protein Loadings using the getPathPCLs()
function:
wp195PCLs_ls <- getPathPCLs(ovarian_aespcOut, "WP195") wp195PCLs_ls
The proteins with non-zero loadings can be extracted as follows:
wp195Loadings_df <- wp195PCLs_ls$Loadings %>% filter(PC1 != 0) wp195Loadings_df
We can also prepare these loadings for graphics:
wp195Loadings_df <- wp195Loadings_df %>% # Sort Loading from Strongest to Weakest arrange(desc(abs(PC1))) %>% # Order the Genes by Loading Strength mutate(featureID = factor(featureID, levels = featureID)) %>% # Add Directional Indicator (for Colour) mutate(Direction = factor(ifelse(PC1 > 0, "Up", "Down")))
Now we will construct a column chart with ggplot2
's geom_col()
function.
ggplot(data = wp195Loadings_df) + # Set overall appearance theme_bw() + # Define the dependent and independent variables aes(x = featureID, y = PC1, fill = Direction) + # From the defined variables, create a vertical bar chart geom_col(width = 0.5, fill = "#005030", color = "#f47321") + # Set main and axis titles labs( title = "Gene Loadings on IL-1 Signaling Pathway", x = "Protein", y = "Loadings of PC1" ) + # Remove the legend guides(fill = FALSE)
Alternatively, we can also plot the correlation of each gene with first PC for each gene. These correlations can be computed by using the TidyCorrelation()
function in pathwayPCA
's vignette “Chapter 5 - Visualizing the Results”, Section 3.3.^[https://gabrielodom.github.io/pathwayPCA/articles/Supplement5-Analyse_Results.html#gene-correlations-with-pcs]
In the study of complex diseases, there is often considerable heterogeneity among different subjects with regard to underlying causes of disease and benefit of particular treatment. Therefore, in addition to identifying disease-relevant pathways for the entire patient group, successful (personalized) treatment regimens will also depend upon knowing if a particular pathway is dysregulated for an individual patient.
To this end, we can also assess subject-specific pathway activity. As we saw earlier, the getPathPCLs()
function also returns subject-specific estimates for the individual pathway PCs. We can plot these as follows.
ggplot(data = wp195PCLs_ls$PCs) + # Set overall appearance theme_classic() + # Define the independent variable aes(x = V1) + # Add the histogram layer geom_histogram(bins = 10, color = "#005030", fill = "#f47321") + # Set main and axis titles labs( title = "Distribution of Sample-specific Estimate of Pathway Activities", subtitle = paste0(wp195PCLs_ls$pathway, ": ", wp195PCLs_ls$description), x = "PC1 Score for Each Sample", y = "Count" )
This graph shows there can be considerable heterogeneity in pathway activities between the patients.
Users are often also interested in examining the actual data used for analysis of the top pathways, especially for the relevant genes with the pathway. To extract this dataset, we can use the SubsetPathwayData()
function. These commands extract data for the IL-1 signaling pathway:
wp195Data_df <- SubsetPathwayData(ov_OmicsSurv, "WP195") wp195Data_df
We can also perform analysis for individual genes belonging to the pathway:
library(survival) NFKB1_df <- wp195Data_df %>% select(EventTime, EventObs, NFKB1) wp195_mod <- coxph( Surv(EventTime, EventObs) ~ NFKB1, data = NFKB1_df )
Now we inspect the output from the Cox PH model (we've included the pretty version using the prettify()
function from the papeR
package^[https://CRAN.R-project.org/package=papeR]):
summary(wp195_mod)
wp195pretty_mod <- prettify( summary(wp195_mod) ) kable(wp195pretty_mod)
Additionally, we can estimate Kaplan-Meier survival curves for patients with high or low expression values for individual genes:
NFKB1_df <- NFKB1_df %>% # Group subjects by gene expression mutate(NFKB1_Expr = ifelse(NFKB1 > median(NFKB1), "High", "Low")) %>% # Re-code time to years mutate(EventTime = EventTime / 365.25) %>% # Ignore any events past 10 years filter(EventTime <= 10) # Fit the survival model NFKB1_fit <- survfit( Surv(EventTime, EventObs) ~ NFKB1_Expr, data = NFKB1_df )
Finally, we can plot these K-M curves over NFKB1 protein expression.
ggsurvplot( NFKB1_fit, # No confidence intervals; add the p-value conf.int = FALSE, pval = TRUE, # Show times to median survival surv.median.line = "hv", xlab = "Time in Years", palette = c("#f47321", "#005030") )
While copy number alterations are common genomic aberrations in ovarian carcer, recent studies have shown these changes do not necessarily lead to concordant changes in protein expression. In Section 1.5.3 above, we illustrated testing pathway activities in protein expression against survival outcome. In this section, we will additionally test pathway activities in copy number against survival outcome. Moreover, we will perform integrative analysis to identify those survival-associated protein pathways, genes, and samples driven by copy number alterations.
Omics
data objectWe can identify copy number (CNV) pathways significantly associated with survival in the same way as we did for protein expressions. This gene-level CNV data was downloaded from the same LinkedOmics database.
data("ovCNV_df") ovCNV_df[, 1:5]
And now we create an Omics
data container. (Note: because these analysis steps take a little longer than the steps shown previously--3 minutes over 20 cores, we include the AESPCA output directly. You don't need to execute the code in the next two chunks.)
# DONT RUN ovCNV_Surv <- CreateOmics( assayData_df = ovCNV_df, pathwayCollection_ls = wikipathways_PC, response = ovPheno_df, respType = "survival", minPathSize = 5 )
1230 gene name(s) are invalid. Invalid name(s) are: ... There are 549 samples shared by the assay and phenotype data. ====== Creating object of class OmicsSurv ======= The input pathway database included 5831 unique features. The input assay dataset included 24776 features. Only pathways with at least 5 or more features included in the assay dataset are tested (specified by minPathSize parameter). There are 424 pathways which meet this criterion. Because pathwayPCA is a self-contained test (PMID: 17303618), only features in both assay data and pathway database are considered for analysis. There are 5637 such features shared by the input assay and pathway database.
Finally, we can apply the AESPCA method to this copy-number data container. Due to the large sample size, this will take a few moments.
# DONT RUN ovCNV_aespcOut <- AESPCA_pVals( object = ovCNV_Surv, numPCs = 1, parallel = TRUE, numCores = 20, numReps = 0, adjustment = "BH" )
Part 1: Calculate Pathway AES-PCs Initializing Computing Cluster: DONE Extracting Pathway PCs in Parallel: DONE Part 2: Calculate Pathway p-Values Initializing Computing Cluster: DONE Extracting Pathway p-Values in Parallel: DONE Part 3: Adjusting p-Values and Sorting Pathway p-Value Data Frame DONE
Rather than execute this code yourself, we have included the output object with this package:
data(ovCNV_aespcOut)
Next, we identify the intersection of significant pathways based on both CNV and protein data. First, we will create a data frame of the pathway $p$-values from both CNV and proteomics.
# Copy Number CNVpvals_df <- getPathpVals(ovCNV_aespcOut, alpha = 0.05) %>% mutate(rawp_CNV = rawp) %>% select(description, rawp_CNV) # Proteomics PROTpvals_df <- getPathpVals(ovarian_aespcOut, alpha = 0.05) %>% mutate(rawp_PROT = rawp) %>% select(description, rawp_PROT) # Intersection SigBoth_df <- inner_join(PROTpvals_df, CNVpvals_df, by = "description") # WnT Signaling Pathway is listed as WP363 and WP428
The results showed there are 32 pathways significantly associated with survival in both CNV and protein data, which is significantly more than expected by chance (p-value = 0.00065; Fisher's Exact Test; shown in multi_pathway_overlap_fishers.R
). Here are the top-10 most significant pathways (sorted by protein data significance):
SigBoth_df[1:10, ]
kable( SigBoth_df[1:10, ] )
Similar to the protein pathway analysis shown in Section 1.5.3, we can also identify relevant genes with nonzero loadings that drives pathway significance in CNV. The "IL-1 signaling pathway" (WP195) is significant in both CNV and protein data, so we look for genes with non-zero loadings in both.
# Copy Number Loadings CNVwp195_ls <- getPathPCLs(ovCNV_aespcOut, "WP195") CNV195load_df <- CNVwp195_ls$Loadings %>% filter(PC1 != 0) %>% rename(PC1_CNV = PC1) # Protein Loadings PROTwp195_ls <- getPathPCLs(ovarian_aespcOut, "WP195") PROT195load_df <- PROTwp195_ls$Loadings %>% filter(PC1 != 0) %>% rename(PC1_PROT = PC1) # Intersection inner_join(CNV195load_df, PROT195load_df)
The result showed that NFKB1, IKBKB, and other genes are selected by AESPCA when testing IL-1 signaling pathway (WP195) against survival outcome in both CNV and protein pathway analysis.
Also in Section 1.5.3, we have seen that there can be considerable heterogeneity in pathway activities between patients. One possible reason could be that copy number changes might not directly result in changes in protein expression for some of the patients. pathwayPCA
can be used to estimate pathway activities for each patient, for copy number, gene, and protein expressions separately. These estimates can then be viewed jointly using a Circos plot.
include_graphics(file.path(figDir_char, "circosPlot_20190617.png"))
The accompanying Circos plot shown normalized copy number (inner circle) and protein expression (outer circle) pathway activities for the IL-1 signaling pathway in the ovarian cancer dataset samples. Each bar corresponds to a patient sample. Red bars indicate lower expression values and negative pathway activity for the sample, while blue color bars indicate higher expression values and stronger pathway activity for the sample. Note that only some patients have concordant changes in copy number and protein expression. The code to make this plot is included in inst/scripts/circle_plot.R
.
pathwayPCA
is capable of analyzing studies with multiple experimental factors. In this section, we illustrate using pathwayPCA
to test differential association of pathway expression with survival outcome in male and female subjects.
For this example, we used IlluminaHiSeq gene expression RNAseq data from the TCGA KIRP cohort, which we downloaded from the Xena browser datahub.^[https://xenabrowser.net/datapages/] (This process is described in inst/scripts/explore_KIRP.R
.) First, we load the KIRP data, create an Omics
data container, and extract first AESPC from each pathway.
data("kirpPheno_df") data("kirpRNAseq_df")
Once again, this will take a few moments (on my machine, this takes 1.9 minutes over 20 cores). (You can skip the next two chunks and load the results directly.)
# DONT RUN kidney_Omics <- CreateOmics( assayData_df = kirpRNAseq_df, pathwayCollection_ls = wikipathways_PC, # Exclude the gender indicator for now response = kirpPheno_df[, 1:3], respType = "surv", minPathSize = 5 )
317 genes have variance < epsilon and will be removed. These gene(s) are: ... 30 gene name(s) are invalid. Invalid name(s) are: ... There are 320 samples shared by the assay and phenotype data. ====== Creating object of class OmicsSurv ======= The input pathway database included 5831 unique features. The input assay dataset included 20213 features. Only pathways with at least 5 or more features included in the assay dataset are tested (specified by minPathSize parameter). There are 423 pathways which meet this criterion. Because pathwayPCA is a self-contained test (PMID: 17303618), only features in both assay data and pathway database are considered for analysis. There are 5566 such features shared by the input assay and pathway database.
# DONT RUN kirpRNAseq_aespcOut <- AESPCA_pVals( object = kidney_Omics, numPCs = 1, parallel = TRUE, numCores = 20, numReps = 0, adjustment = "BH" )
Part 1: Calculate Pathway AES-PCs Initializing Computing Cluster: DONE Extracting Pathway PCs in Parallel: DONE Part 2: Calculate Pathway p-Values Initializing Computing Cluster: DONE Extracting Pathway p-Values in Parallel: DONE Part 3: Adjusting p-Values and Sorting Pathway p-Value Data Frame DONE
Rather than execute this code yourself, we have included the output object with this package:
data("kirpRNAseq_aespcOut")
Next we can test whether there is differential pathway association with survival outcome for males and females by the following model, [ h(t) = h_0(t) \exp\left[ \beta_1\text{PC}_1 + \beta_2\text{male} + \beta_3(\text{PC}_1 \times \text{male}) \right]. ]
In this model, $h(t)$ is expected hazard at time $t$, $h_0 (t)$ is baseline hazard when all predictors are zero, variable $male$ is an indicator variable for male samples, and $PC_1$ is a pathway’s estimated first principal component based on AESPCA.
In order to test the sex interaction effect for all pathways, we will write a function which tests the interaction effect for one pathway.
TestIntxn <- function(pathway, pcaOut, resp_df){ # For the given pathway, extract the PCs and loadings from the pcaOut list PCL_ls <- getPathPCLs(pcaOut, pathway) # Select and rename the PC PC_df <- PCL_ls$PCs %>% rename(PC1 = V1) # Join the phenotype data to this PC data_df <- inner_join(resp_df, PC_df, by = c("Sample" = "sampleID")) # Construct a survival model with sex interaction sex_mod <- coxph( Surv(time, status) ~ PC1 + male + PC1 * male, data = data_df ) # Extract the model fit statistics for the interaction modStats_mat <- t( coef(summary(sex_mod))["PC1:maleTRUE", ] ) colnames(modStats_mat) <- c("coef", "exp_coef", "se_coef", "z", "pVal") # Return a data frame of the pathway and model statistics list( statistics = data.frame( terms = pathway, description = PCL_ls$description, modStats_mat, stringsAsFactors = FALSE ), model = sex_mod ) }
As an example, we can test it on pathway WP195,
TestIntxn("WP195", kirpRNAseq_aespcOut, kirpPheno_df)$model
We can also apply this function to our list of pathways.
# List of pathways paths_char <- kirpRNAseq_aespcOut$pVals_df$terms # Apply over this list interactions_ls <- sapply( paths_char, FUN = TestIntxn, pcaOut = kirpRNAseq_aespcOut, resp_df = kirpPheno_df, simplify = FALSE )
We will tabulate this list and sort the pathways by significance:
interactions_df <- # Take list of interactions interactions_ls %>% # select the first element (the data frame of model stats) lapply(`[[`, 1) %>% # stack these data frames bind_rows() %>% as_tibble() %>% # sort the rows by significance arrange(pVal)
Now, we can inspect the pathways with significant gender-PC1 interaction (note that we have not adjusted these pathway results for false discovery rate; these examples are only to show functionality of the package).
interactions_df %>% filter(pVal < 0.05)
kable( interactions_df %>% filter(pVal < 0.05) )
The results showed the most significant pathway is WP1559 (“TFs Regulate miRNAs related to cardiac hypertrophy”). We can inspect the model results for this pathway directly.
summary(interactions_ls[["WP1559"]]$model)
The results showed that although sex is not significantly associated with survival outcome, the association of pathway gene expression (PC1) with survival is highly dependent on sex of the samples.
For visualization, we first match the PC expression to survival outcome.
# Bind the Pheno Data to WP1559's First PC kidneyWP1559_df <- inner_join( kirpPheno_df, getPathPCLs(kirpRNAseq_aespcOut, "WP1559")$PCs, by = c("Sample" = "sampleID") )
Now we divide subjects according to sex and high or low PC-expression, and add a group indicator for the four groups.
# Add Grouping Feature kidneySurvWP1559grouped_df <- kidneyWP1559_df %>% rename(PC1 = V1) %>% # add strength indicator mutate(direction = ifelse(PC1 > median(PC1), "High", "Low")) %>% # group by interaction of sex and strength on PC mutate(group = paste0(direction, ifelse(male, "_Male", "_Female"))) %>% # recode time in years mutate(time = time / 365.25) %>% # remove summarized columns select(-male, -PC1, -direction)
Now we can plot survival curves for the four groups.
# Fit the survival model fit <- survfit( Surv(time, status) ~ group, data = kidneySurvWP1559grouped_df ) ggsurvplot( fit, conf.int = FALSE, xlab = "Time in Years", ylim = c(0.4, 1), xlim = c(0, 10) )
These Kaplan-Meier curves show that while high or low pathway activities were not significantly associated with survival in male subjects (green and purple curves, respectively), female subjects with high pathway activities (red) had significantly worse survival outcomes than those with low pathway activities (blue).
For addtional information on pathwayPCA
, each of our supplementary vignette chapters contain detailed tutorials on the topics discussed above. These vignettes are:
Omics
Data ObjectsChen, X., Wang, L., Smith, J.D. and Zhang, B. (2008) Supervised principal component analysis for gene set enrichment of microarray data with continuous or survival outcomes. Bioinformatics, 24, 2474-2481.
Chen, X., Wang, L., Hu, B., Guo, M., Barnard, J. and Zhu, X. (2010) Pathway-based analysis for genome-wide association studies using supervised principal components. Genetic epidemiology, 34, 716-724.
Chen, X. (2011) Adaptive elastic-net sparse principal component analysis for pathway association testing. Statistical applications in genetics and molecular biology, 10.
The R session information for this vignette is:
sessionInfo()
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.