if (!require("BiocManager"))
install.packages("BiocManager")
BiocManager::install("glmSparseNet")
library(dplyr)
library(ggplot2)
library(survival)
library(loose.rock)
library(futile.logger)
library(curatedTCGAData)
library(TCGAutils)
#
library(glmSparseNet)
#
# Some general options for futile.logger the debugging package
.Last.value <- flog.layout(layout.format('[~l] ~m'))
.Last.value <- loose.rock::show.message(FALSE)
# Setting ggplot2 default theme as minimal
theme_set(ggplot2::theme_minimal())
The data is loaded from an online curated dataset downloaded from TCGA using
curatedTCGAData
bioconductor package and processed.
To accelerate the process we use a very reduced dataset down to 107 variables only (genes), which is stored as a data object in this package. However, the procedure to obtain the data manually is described in the following chunk.
brca <- curatedTCGAData(diseaseCode = "BRCA", assays = "RNASeq2GeneNorm", FALSE)
brca <- TCGAutils::splitAssays(brca, c('01','11'))
xdata.raw <- t(cbind(assay(brca[[1]]), assay(brca[[2]])))
# Get matches between survival and assay data
class.v <- TCGAbiospec(rownames(xdata.raw))$sample_definition %>% factor
names(class.v) <- rownames(xdata.raw)
# keep features with standard deviation > 0
xdata.raw <- xdata.raw %>%
{ (apply(., 2, sd) != 0) } %>%
{ xdata.raw[, .] } %>%
scale
set.seed(params$seed)
small.subset <- c('CD5', 'CSF2RB', 'HSF1', 'IRGC', 'LRRC37A6P', 'NEUROG2',
'NLRC4', 'PDE11A', 'PIK3CB', 'QARS', 'RPGRIP1L', 'SDC1',
'TMEM31', 'YME1L1', 'ZBTB11',
sample(colnames(xdata.raw), 100))
xdata <- xdata.raw[, small.subset[small.subset %in% colnames(xdata.raw)]]
ydata <- class.v
Fit model model penalizing by the hubs using the cross-validation function by
cv.glmHub
.
fitted <- cv.glmHub(xdata, ydata,
family = 'binomial',
network = 'correlation',
nlambda = 1000,
network.options = networkOptions(cutoff = .6,
min.degree = .2))
Shows the results of 1000
different parameters used to find the optimal value
in 10-fold cross-validation. The two vertical dotted lines represent the best
model and a model with less variables selected (genes), but within a standard
error distance from the best.
plot(fitted)
Taking the best model described by lambda.min
coefs.v <- coef(fitted, s = 'lambda.min')[,1] %>% { .[. != 0]}
coefs.v %>% {
data.frame(ensembl.id = names(.),
gene.name = geneNames(names(.))$external_gene_name,
coefficient = .,
stringsAsFactors = FALSE)
} %>%
arrange(gene.name) %>%
knitr::kable()
ensembl.id | gene.name | coefficient | |
---|---|---|---|
(Intercept) | (Intercept) | (Intercept) | -6.8189811 |
CD5 | CD5 | AMOTL1 | -1.1200445 |
NLRC4 | NLRC4 | ATR | -1.4434577 |
PIK3CB | PIK3CB | B3GALT2 | -0.3880002 |
ZBTB11 | ZBTB11 | BAG2 | -0.3325728 |
ATR | ATR | C16orf82 | 1.2498303 |
IL2 | IL2 | CD5 | 0.6327083 |
GDF11 | GDF11 | CIITA | -0.2676642 |
DCP1A | DCP1A | DCP1A | 0.2994599 |
AMOTL1 | AMOTL1 | FAM86B1 | 0.4430643 |
BAG2 | BAG2 | FNIP2 | -0.1841676 |
C16orf82 | C16orf82 | GDF11 | 0.0396368 |
FAM86B1 | FAM86B1 | GNG11 | 0.2025462 |
FNIP2 | FNIP2 | GREM2 | 0.6101758 |
MS4A4A | MS4A4A | GZMB | 1.1614778 |
B3GALT2 | B3GALT2 | HAX1 | -0.0867011 |
GNG11 | GNG11 | IL2 | 3.0659065 |
NDRG2 | NDRG2 | MMP28 | 1.1142519 |
HAX1 | HAX1 | MS4A4A | -0.1516836 |
GREM2 | GREM2 | NDRG2 | -0.2014884 |
CIITA | CIITA | NLRC4 | 0.4256103 |
GZMB | GZMB | PIK3CB | -2.7663573 |
MMP28 | MMP28 | ZBTB11 | -0.8438023 |
geneNames(names(coefs.v)) %>% { hallmarks(.$external_gene_name)$heatmap }
## [INFO] Misclassified (11)
## [INFO] * False primary solid tumour: 7
## [INFO] * False normal : 4
Histogram of predicted response
ROC curve
## Setting levels: control = Primary Solid Tumor, case = Solid Tissue Normal
## Setting direction: controls < cases