Modality Graph Generation
Introduction
This R Markdown file generates K Nearest Neighbour graphs for different modalities in a given dataset.
Setup - Insert Libraries
source('~/MOGDx/R/preprocess_functions.R')
Load Data, Specify Locations, Projects and Modalities
trait = 'paper_BRCA_Subtype_PAM50'
dataset = 'TCGA'
project = 'BRCA'
for (modality in c( 'mRNA' )) {
print(modality)
load(paste0('./../data/',dataset , '/raw/', project ,'/',modality,'_processed.RData'))
if (modality %in% c('miRNA' , 'mRNA')) {
g <- expr.to.graph(datExpr , datMeta , trait , top_genes , modality)
} else if (modality == 'DNAm') {
g <- expr.to.graph(datExpr , datMeta , trait , cpg_sites , modality)
} else if (modality == 'CNV') {
g <- expr.to.graph(log(datExpr) , datMeta , trait , cnv_sites , modality)
} else if (modality == 'RPPA') {
g <- expr.to.graph(datExpr , datMeta , trait , protein_sites , modality)
} else if (modality == 'CSF') {
g <- expr.to.graph(datExpr , datMeta , trait , csf_sites , modality)
} else if (modality == 'MOCA') {
g <- expr.to.graph(datExpr , datMeta , trait , q_sites , modality)
} else if (modality == 'MDS-UPDRS') {
g <- expr.to.graph(datExpr , datMeta , trait , p_sites , modality)
} else if (modality == 'Parkinsonism') {
g <- expr.to.graph(datExpr , datMeta , trait , park_sites , modality)
} else if (modality == 'Clinical') {
g <- expr.to.graph(datExpr , datMeta , trait , clin_sites , modality)
} else if (modality == 'SNP') {
g <- expr.to.graph(datExpr , datMeta , trait , snp_sites , modality)
}
write.csv(g, file = paste0('./../data/',dataset , '/raw/', project , '/output/',modality,'_graph.csv'))
write.csv(datExpr , file = paste0('./../data/',dataset , '/raw/', project , '/output','/datExpr_', modality , '.csv'))
write.csv(datMeta , file = paste0('./../data/',dataset , '/raw/', project , '/output','/datMeta_', modality , '.csv'))
}
## [1] "mRNA"
## [1] 2484 1047
