if (!require("BiocManager")) { install.packages("BiocManager") } BiocManager::install("glmSparseNet")
library(dplyr) library(Matrix) library(ggplot2) library(forcats) library(parallel) library(reshape2) library(survival) library(VennDiagram) library(futile.logger) library(curatedTCGAData) library(MultiAssayExperiment) library(TCGAutils) # library(glmSparseNet) # # # Some general options for futile.logger the debugging package flog.layout(layout.format("[~l] ~m")) options( "glmSparseNet.show_message" = FALSE, "glmSparseNet.base_dir" = withr::local_tempdir() ) # Setting ggplot2 default theme as minimal theme_set(ggplot2::theme_minimal())
This vignette uses the STRING database (https://string-db.org/) of protein-protein interactions as the network-based penalizer in generalized linear models using Breast invasive carcinoma sample dataset.
The degree vector is calculated manually to account for genes that are not present in the STRING database, as these will not have any interactions, i.e. edges.
Retrieve all interactions from STRING databse. We have included a helper function that retrieves the Homo sapiens known interactions.
For this vignette, we use a cached version of all interaction with
score_threshold = 700
Note: Text-based interactions are excluded from the network.
# Not evaluated in vignette as it takes too long to download and process allInteractions700 <- stringDBhomoSapiens(scoreThreshold = 700) stringNetwork <- buildStringNetwork(allInteractions700, "external")
data("string.network.700.cache", package = "glmSparseNet") stringNetwork <- string.network.700.cache
Build a sparse matrix object that contains the network.
stringNetworkUndirected <- stringNetwork + Matrix::t(stringNetwork) stringNetworkUndirected <- (stringNetworkUndirected != 0) * 1
flog.info("Directed graph (score_threshold = %d)", 700) flog.info(" * total edges: %d", sum(stringNetwork != 0)) flog.info(" * unique protein: %d", nrow(stringNetwork)) flog.info( " * edges per protein: %f", sum(stringNetwork != 0) / nrow(stringNetwork) ) flog.info("") flog.info("Undirected graph (score_threshold = %d)", 700) flog.info(" * total edges: %d", sum(stringNetworkUndirected != 0) / 2) flog.info(" * unique protein: %d", nrow(stringNetworkUndirected)) flog.info( " * edges per protein: %f", sum(stringNetworkUndirected != 0) / 2 / nrow(stringNetworkUndirected) )
stringNetworkBinary <- (stringNetworkUndirected != 0) * 1 degreeNetworkVector <- ( Matrix::rowSums(stringNetworkBinary) + Matrix::colSums(stringNetworkBinary) ) / 2 flog.info("Summary of degree:", summary(degreeNetworkVector), capture = TRUE)
qplot( degreeNetworkVector[ degreeNetworkVector <= quantile(degreeNetworkVector, probs = .99999) ], geom = "histogram", fill = my.colors(2), bins = 100 ) + theme(legend.position = "none") + xlab("Degree (up until 99.999% quantile)")
glmSparseNet
# chunk not included as it produces to many unnecessary messages brca <- tryCatch( { curatedTCGAData( diseaseCode = "BRCA", assays = "RNASeq2GeneNorm", version = "1.1.38", dry.run = FALSE ) }, error = function(err) { NULL } )
brca <- curatedTCGAData( diseaseCode = "BRCA", assays = "RNASeq2GeneNorm", version = "1.1.38", dry.run = FALSE )
Build the survival data from the clinical columns.
xdata
and ydata
# keep only solid tumour (code: 01) brcaPrimarySolidTumor <- TCGAutils::TCGAsplitAssays(brca, "01") xdataRaw <- t(assay(brcaPrimarySolidTumor[[1]])) # Get survival information ydataRaw <- colData(brcaPrimarySolidTumor) |> as.data.frame() |> # Convert days to integer dplyr::mutate( Days.to.date.of.Death = as.integer(Days.to.date.of.Death), Days.to.Last.Contact = as.integer(Days.to.Date.of.Last.Contact) ) |> # Find max time between all days (ignoring missings) dplyr::rowwise() |> dplyr::mutate( time = max(days_to_last_followup, Days.to.date.of.Death, Days.to.Last.Contact, days_to_death, na.rm = TRUE ) ) |> # Keep only survival variables and codes dplyr::select(patientID, status = vital_status, time) |> # Discard individuals with survival time less or equal to 0 dplyr::filter(!is.na(time) & time > 0) |> as.data.frame() # Set index as the patientID rownames(ydataRaw) <- ydataRaw$patientID # keep only features that are in degreeNetworkVector and have # standard deviation > 0 validFeatures <- colnames(xdataRaw)[ colnames(xdataRaw) %in% names(degreeNetworkVector[degreeNetworkVector > 0]) ] xdataRaw <- xdataRaw[ TCGAbarcode(rownames(xdataRaw)) %in% rownames(ydataRaw), validFeatures ] xdataRaw <- scale(xdataRaw) # Order ydata the same as assay ydataRaw <- ydataRaw[TCGAbarcode(rownames(xdataRaw)), ] # Using only a subset of genes previously selected to keep this short example. set.seed(params$seed) smallSubset <- c( "AAK1", "ADRB1", "AK7", "ALK", "APOBEC3F", "ARID1B", "BAMBI", "BRAF", "BTG1", "CACNG8", "CASP12", "CD5", "CDA", "CEP72", "CPD", "CSF2RB", "CSN3", "DCT", "DLG3", "DLL3", "DPP4", "DSG1", "EDA2R", "ERP27", "EXD1", "GABBR2", "GADD45A", "GBP1", "HTR1F", "IFNK", "IRF2", "IYD", "KCNJ11", "KRTAP5-6", "MAFA", "MAGEB4", "MAP2K6", "MCTS1", "MMP15", "MMP9", "NFKBIA", "NLRC4", "NT5C1A", "OPN4", "OR13C5", "OR13C8", "OR2T6", "OR4K2", "OR52E6", "OR5D14", "OR5H1", "OR6C4", "OR7A17", "OR8J3", "OSBPL1A", "PAK6", "PDE11A", "PELO", "PGK1", "PIK3CB", "PMAIP1", "POLR2B", "POP1", "PPFIA3", "PSME1", "PSME2", "PTEN", "PTGES3", "QARS", "RABGAP1", "RBM3", "RFC3", "RGPD8", "RPGRIP1L", "SAV1", "SDC1", "SDC3", "SEC16B", "SFPQ", "SFRP5", "SIPA1L1", "SLC2A14", "SLC6A9", "SPATA5L1", "SPINT1", "STAR", "STXBP5", "SUN3", "TACC2", "TACR1", "TAGLN2", "THPO", "TNIP1", "TP53", "TRMT2B", "TUBB1", "VDAC1", "VSIG8", "WNT3A", "WWOX", "XRCC4", "YME1L1", "ZBTB11", "ZSCAN21" ) |> sample(size = 50) |> sort() # make sure we have 100 genes smallSubset <- c(smallSubset, sample(colnames(xdataRaw), 51)) |> unique() |> sort() xdata <- xdataRaw[, smallSubset[smallSubset %in% colnames(xdataRaw)]] ydata <- ydataRaw |> dplyr::select(time, status) |> dplyr::filter(!is.na(time) | time < 0)
# # Add degree 0 to genes not in STRING network myDegree <- degreeNetworkVector[smallSubset] myString <- stringNetworkBinary[smallSubset, smallSubset]
Degree distribution for sample set of gene features (in xdata).
qplot(myDegree, bins = 100, fill = my.colors(3)) + theme(legend.position = "none")
set.seed(params$seed) foldid <- glmSparseNet:::balancedCvFolds(ydata$status)$output
# List that will store all selected genes selectedGenes <- list()
Penalizes using the Hub heuristics, see hubHeuristic
function definition for
more details.
resultCVHub <- cv.glmHub(xdata, Surv(ydata$time, ydata$status), family = "cox", foldid = foldid, network = myString, network.options = networkOptions(minDegree = 0.2) )
Kaplan-Meier estimator separating individuals by low and high risk (based on model's coefficients)
separate2GroupsCox( as.vector(coef(resultCVHub, s = "lambda.min")[, 1]), xdata, ydata, plot.title = "Full dataset", legend.outside = FALSE ) selectedGenes[["Hub"]] <- Filter( function(.x) .x != 0, coef(resultCVHub, s = "lambda.min")[, 1] ) |> names() |> geneNames() |> magrittr::extract2("external_gene_name")
Penalizes using the Orphan heuristics, see orphanHeuristic
function
definition for more details.
resultCVOrphan <- cv.glmOrphan(xdata, Surv(ydata$time, ydata$status), family = "cox", foldid = foldid, network = myString, network.options = networkOptions(minDegree = 0.2) )
Kaplan-Meier estimator separating individuals by low and high risk (based on model's coefficients)
separate2GroupsCox( as.vector(coef(resultCVOrphan, s = "lambda.min")[, 1]), xdata, ydata, plot.title = "Full dataset", legend.outside = FALSE ) selectedGenes[["Orphan"]] <- Filter( function(.x) .x != 0, coef(resultCVOrphan, s = "lambda.min")[, 1] ) |> names() |> geneNames() |> magrittr::extract2("external_gene_name")
Uses regular glmnet model as simple baseline
library(glmnet) resultCVGlmnet <- cv.glmnet(xdata, Surv(ydata$time, ydata$status), family = "cox", foldid = foldid )
Kaplan-Meier estimator separating individuals by low and high risk (based on model's coefficients)
separate2GroupsCox( as.vector(coef(resultCVGlmnet, s = "lambda.min")[, 1]), xdata, ydata, plotTitle = "Full dataset", legendOutside = FALSE ) selectedGenes[["GLMnet"]] <- Filter( function(.x) .x != 0, coef(resultCVGlmnet, s = "lambda.min")[, 1] ) |> names() |> geneNames() |> magrittr::extract2("external_gene_name")
Venn diagram of overlapping genes.
vennPlot <- venn.diagram( selectedGenes, NULL, fill = c( myColors(5), myColors(3), myColors(4) ), alpha = c(0.3, 0.3, .3), cex = 2, cat.fontface = 4, category.names = names(selectedGenes) ) grid.draw(vennPlot)
Descriptive table showing which genes are selected in each model
We can observe, that elastic net without network-based penalization selects the best model with 40% more genes than glmOrphan and glmHub, without loosing accuracy.
note: size of circles represent the degree of that gene in network.
melt(selectedGenes) |> mutate( Degree = myDegree[value], value = factor(value), L1 = factor(L1) ) |> mutate(value = fct_reorder(value, Degree)) |> as.data.frame() |> ggplot() + geom_point( aes(value, L1, size = Degree), shape = mySymbols(3), color = myColors(3) ) + theme( axis.text.x = element_text(angle = 90, hjust = 1, vjust = 0), legend.position = "top" ) + ylab("Model") + xlab("Gene") + scale_size_continuous(trans = "log10")
sessionInfo()
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.