In JiantaoShi/NetORA: Network-based pathway over-representation analysis (NetORA)

Intruduction

Genome wide association studies (GWAS) have been successfully used to identify disease-associated variants, however the causal genes in many diseases remain elusive, due to effects such as linkage disequilibrium (LD) between associated variants and long-range regulation. Direct experimental validation of the many potential causal genes is expensive and difficult, so an attractive first step is to prioritize genes with respect their biological relevance. Numerous evidences suggest genes function in pathway-level, so different disease causal genes might function in the same causal pathway. Pathways can be annotated as un-structured gene sets (Reactome), structured tree (GO) or even networks. We have implemented the algorithm of GRAIL [@Raychaudhuri2009] in R as a package NetGPA and provide both text-mining and coexpression networks. We have demonstrated coexpression network-based gene prioritization is more sensitive when gene expression signatures are provided as seeds [@Roccaro2016]. NetORA is an extension of NetGPA to perform network-based pathway enrichment analysis.

Standard workflow

To demonstrate the input and output data format in NetORA, we use example data sets in NetGPA.

library("NetGPA")
library("NetORA")
data("Example_NetGPA")

names(Example_NetGPA)

Input

Similar as other pathway-enrichment analysis, NetORA expect annotated pathways and signatures as input. Since it's a network-based method, a network is also required.

Pathways

In vignettes of NetGPA, we have used gene prioritization to identify pathways that drive DNA methylation transformation [@Smith2017]. Here we took an more intuitive approach by directly performing network-based pathway enrichment analysis.

# Genes near hyper-methylated CpG Islands in Mouse ExE
ExE_Hyper     <- Example_NetGPA$ExE_Hyper

We treat ExE_Hyper as an annotated pathway.

Signatures

Signatures are gene sets we want to test. Here we use pathways that are frequently mutated in cancers as signatures.

# show example query genes
Cancer_GeneSet   <- Example_NetGPA$Cancer_GeneSet
names(Cancer_GeneSet)

Background

Background represents all possible genes evaluated when signatures are defined. For microarray studies, background is all genes measured in a microarray platform; For RNA-seq studies, background is all genes in genome.

Global networks

NetORA us NetGPA a back-end and thus requires networks in the same format, as described below. A gene network is represented as an integer matrix, in which column names are all genes included and each column contains top nearest neighbors of the gene indicated by column name. Here we will use a global text-mining network as an example.

# build a example global gene-network
data(text_2006_12_NetGPA)
networkMatrix <- text_2006_12_NetGPA

dim(networkMatrix)
networkMatrix[1:10, c("IL12B", "TET2")]
colnames(networkMatrix)[7408]

In the example shown above, a network covers 18835 genes and the nearest neighbor of IL12B is shown as 7408, which is the 7408th element of column names (gene IL12A).

Quick Start

Now we have a pathway ExE_Hyper, gene signatures in Cancer_GeneSet and networks in text_2006_12_NetGPA.

# Network-based pathway enrichment analysis
queryTable <- NetORA_Pre(ExE_Hyper, text_2006_12_NetGPA, progressBar = FALSE)
PG <- colnames(text_2006_12_NetGPA)
mergedT <- NetORA_GS(Cancer_GeneSet, queryTable, PG, FDR = 0.05)

mergedT[order(mergedT$pvalue), ]

You can clearly see that only FGF-related signaling pathways are statistically significant.

Pre-computed networks

Building networks is time-consuming, we thus have pre-computed networks for all canonical pathways defined in MSigDB, using FDR 0.05 as cutoff. They can be loaded as a data set in R data(MSigDB_NetORA_GS). Let's revisit the example we discussed above. In our previous study [@Smith2017], we have identified a list of potential pathways that regulate DNA methylation of a signature genes ExE_Hyper. In above section, we built a network using ExE_Hyper, and tested enrichment potential pathways. Here, we will do it in a reverse way.

data(MSigDB_NetORA_GS)
PG <- colnames(text_2006_12_NetGPA)
CancerPathway <- paste0("REACTOME_", Example_NetGPA$CancerPathway)
mergedT <- NetORA_MSigDB_CP(ExE_Hyper, MSigDB_NetORA_GS[CancerPathway], PG)

mergedT[order(mergedT$pvalue), ]

As expected, FGF-related signaling pathways are more statistically significant than others.

Network databases

NetORA could use all networks that are accepted by NetGPA. In current release of NetGPA, we have provided a text-mining network(text_2006_12_NetGPA), a co-expression network(ce_v12_08_NetGPA) and a integrative network(DEPICT_2015_01_NetGPA).

In the future, we will release more networks, including co-expression networks for Mouse and Rat.

Citation

If you use NetORA in published research, please cite NetORA and also [@Raychaudhuri2009].

Session info

sessionInfo()

References

JiantaoShi/NetORA documentation built on May 28, 2019, 12:43 p.m.