# The Gene Symbol to Entrez annotation function #####
# It depends on properly defining an ID_Map and Aliases data frame
# on the main script for preprocessing.
annot = function(expressionDF, ID_Map, Aliases){
# All exprs datasets have "X" as their first column
expressionDF = expressionDF %>% dplyr::rename(probe = X)
Join = ID_Map %>% right_join(expressionDF) %>%
dplyr::select(probe, EntrezGene.ID, HGNC_Official) %>%
dplyr::filter(is.na(EntrezGene.ID) == FALSE) %>% distinct()
Join$EntrezGene.ID = as.numeric(Join$EntrezGene.ID)
Join = Join[order(Join$EntrezGene.ID),]
dupsg = Join[which(duplicated(Join$EntrezGene.ID)),]
dupsg_names = as.character(dupsg$EntrezGene.ID)
dups_entrez = Join[Join$EntrezGene.ID %in% dupsg_names,]
dups_entrez$EntrezGene.ID = as.numeric(dups_entrez$EntrezGene.ID)
dups_entrez = dups_entrez[order(dups_entrez$EntrezGene.ID),]
dupsp = Join[which(duplicated(Join$probe)),]
dupsp_names = as.character(dupsp$probe)
dups_probes = Join[Join$probe %in% dupsp_names,]
dups_probes$EntrezGene.ID = as.numeric(dups_probes$EntrezGene.ID)
dups_probes = dups_probes[order(dups_probes$probe),]
# The full initial set of duplicates is:
full_dups = rbind(dups_entrez, dups_probes)
full_dups$EntrezGene.ID = as.numeric(full_dups$EntrezGene.ID)
full_dups = full_dups[order(full_dups$EntrezGene.ID, full_dups$probe),]
full_dups = full_dups %>% left_join(Aliases, by ="probe") %>% distinct()
full_dups$final_choice = NA
for(i in 1:nrow(full_dups)){
if(is.na(full_dups$HGNC_Symbol[i]) == T && full_dups$HGNC_Official[i] == "Yes"){
full_dups$HGNC_Symbol[i] = full_dups$probe[i]
full_dups$Entrez[i] = full_dups$EntrezGene.ID[i]
}
}
full_dups = full_dups %>% dplyr::filter(is.na(HGNC_Symbol)==F)
# Dealing with duplicates using a complex for loop the philosophy
# of which is:
# 1) If official HGNC match and same Entrez ID:
# put the probe in final_choice
# 2) If non-official HGNC match but there is a yes somewhere for this
# probe: put NA
# 3) If the probe was not an HGNC symbol but an alias
# check if it has been considered anywhere as an HGNC symbol, if not
# then check if there's only one probe matched to this HGNC symbol
# and if yes check that this is the only probe name (redundant). If
# all are true final_choice is that HGNC symbol
# 4) If the probe was not an HGNC symbol but an alias
# check if it has been considered anywhere as an HGNC symbol, if not
# then check if there's only one HGNC symbol matched to this probe
# and if yes check if the HGNC symbol's name is that name (redundant).
# If all are true then put that HGNC symbol as final_choice
# ELSE put NA
for (i in 1:nrow(full_dups)){
if (full_dups$HGNC_Official[i] == "Yes" &&
full_dups$EntrezGene.ID[i] == full_dups$Entrez[i]){
full_dups$final_choice[i] = full_dups$probe[i]
} else if (full_dups$HGNC_Official[i] == "No" &&
length(unique(full_dups$HGNC_Official[full_dups$probe == full_dups$probe[i]])) > 1){
full_dups$final_choice[i] = NA
} else if (full_dups$HGNC_Official[i] == "No" &&
length(unique(full_dups$HGNC_Official[full_dups$probe == full_dups$probe[i]])) == 1 &&
length(unique(full_dups$probe[full_dups$HGNC_Symbol == full_dups$HGNC_Symbol[i]])) == 1 &&
full_dups$EntrezGene.ID[i] == full_dups$Entrez[i]){
full_dups$final_choice[i] = full_dups$HGNC_Symbol[i]
} else if (full_dups$HGNC_Official[i] == "No" &&
length(unique(full_dups$HGNC_Official[full_dups$probe == full_dups$probe[i]])) == 1 &&
length(unique(full_dups$HGNC_Symbol[full_dups$probe == full_dups$probe[i]])) == 1 &&
length(unique(full_dups$probe[full_dups$EntrezGene.ID == full_dups$EntrezGene.ID[i]])) == 1){
full_dups$final_choice[i] = full_dups$HGNC_Symbol[i]
} else {
full_dups$final_choice[i] = NA
}
}
# Check for unmatched probes
nonas_names = as.character(unique(full_dups$probe[which(is.na(full_dups$final_choice) == FALSE)]))
nas_names = as.character(unique(full_dups$probe[which(is.na(full_dups$final_choice) == TRUE)]))
check = nas_names %in% nonas_names
Unmatched1 = as.data.frame(nas_names[which(check == FALSE)])
colnames(Unmatched1) = "Unmatched probes"
# Ignore the final_choice values of probes that have been matched
# elsewhere
for (b in 1:nrow(full_dups)){
if (full_dups$probe[b] %in% nonas_names &&
is.na(full_dups$final_choice[b]) == TRUE){
full_dups$final_choice[b] = "Ignore"
}
}
# Same but using HGNC symbols so that we ignore the HGNC symbols
# that have been matched
for (c in 1:length(full_dups$probe)){
if (full_dups$HGNC_Symbol[c] %in% nonas_names &&
is.na(full_dups$final_choice[c]) == TRUE){
full_dups$final_choice[c] = "Ignore"
}
}
# Find all probes which do not have final_choice and match them to
# the corresponding HGNC symbol if this is unique and not present
# in final choice. Create a match only when EntrezGene.ID and Entrez
# are the same
for (e in 1:length(full_dups$probe)){
if (is.na(full_dups$final_choice[e]) == TRUE &&
full_dups$HGNC_Official[e] == "No" &&
length(unique(full_dups$HGNC_Official[full_dups$probe == full_dups$probe[e]])) == 1 &&
full_dups$EntrezGene.ID[e] == full_dups$Entrez[e] &&
nrow(distinct(full_dups[full_dups$probe==full_dups$probe[e] &
is.na(full_dups$final_choice) == TRUE,c(1,4)])) == 1){
full_dups$final_choice[e] = full_dups$HGNC_Symbol[e]
}
}
# Check for unmatched probes vol.2
nonas_names = as.character(unique(full_dups$probe[which(is.na(full_dups$final_choice) == FALSE)]))
nas_names = as.character(unique(full_dups$probe[which(is.na(full_dups$final_choice) == TRUE)]))
check = nas_names %in% nonas_names
Unmatched1 = as.data.frame(nas_names[which(check == FALSE)])
colnames(Unmatched1) = "Unmatched probes"
full_dups$final_choice[full_dups$probe %in% Unmatched1$`Unmatched probes`] = "Unclear"
# The most challenging of probes
rogues = full_dups %>% dplyr::filter(final_choice == "Ignore" || final_choice == "Unclear")
rogue_names = unique(rogues$probe)
rights = full_dups %>% dplyr::filter(is.na(final_choice) == FALSE) %>%
dplyr::filter(final_choice != "Ignore") %>%
dplyr::filter(final_choice != "Unclear")
right_names = unique(rights$probe)
real_rogues = rogue_names[which(rogue_names %in% right_names == FALSE)]
bandits = unique(full_dups$probe[full_dups$probe %in% real_rogues]) %>%
as.data.frame()
final_check = Join[Join$probe %in% bandits$.,]
full_dups$EntrezGene.ID[full_dups$probe %in% final_check$probe] = NA
# Filter the full_dups dataset in order to use it to
# correct the Join dataset
full_dups_filt = full_dups %>% dplyr::filter(is.na(final_choice) == FALSE) %>%
dplyr::filter(final_choice != "Unclear") %>%
dplyr::filter(final_choice != "Ignore") %>%
dplyr::select(probe, Entrez, HGNC_Official, final_choice) %>% distinct()
Unmatched2 = as.data.frame(full_dups_filt$probe[which(duplicated(full_dups_filt$probe))])
colnames(Unmatched2) = "Unmatched probes"
Unmatched = unique(rbind(Unmatched1, Unmatched2))
Unmatched_names = unique(Unmatched$`Unmatched probes`)
dup_positions = which(duplicated(full_dups_filt$probe))
dup_names = unique(full_dups_filt$probe[dup_positions])
full_dups_filt$Entrez[full_dups_filt$probe %in% dup_names] = NA
full_dups_filt$Entrez[full_dups_filt$probe %in% final_check$probe] = NA
Join$EntrezGene.ID[Join$probe %in% unique(full_dups$probe)] = "Remove"
Join$EntrezGene.ID[Join$probe %in% Unmatched_names] = "Remove"
Join_pre = Join %>% dplyr::filter(EntrezGene.ID != "Remove") %>%
dplyr::select(-HGNC_Official)
# HGNC to HGNC annotation preparations for intClust
hgnc_annot_filt = full_dups_filt %>% dplyr::select(probe, final_choice) %>%
dplyr::rename(Gene.Symbol = final_choice)
hgnc_annot_pre = Join_pre %>% dplyr::select(probe)
hgnc_annot_pre$Gene.Symbol = as.character(hgnc_annot_pre$probe)
hgnc_annot_pre2 = rbind(hgnc_annot_pre, hgnc_annot_filt)
uniq2 = unique(hgnc_annot_pre2)
hgnc_annot_pre2$Gene.Symbol[hgnc_annot_pre2$probe %in% uniq2$probe[which(duplicated(uniq2$probe) == TRUE)]] = NA
hgnc_annot_pre2 = unique(hgnc_annot_pre2)
hgnc_annot = as.data.frame(expressionDF[,1])
colnames(hgnc_annot) = "probe"
hgnc_annot = hgnc_annot %>% left_join(hgnc_annot_pre2)
# The following is added because CDCA1 and NUF2 create confusions
# in most datasets especially in PAM50
if ("CDCA1" %in% hgnc_annot$probe &&
"NUF2" %in% hgnc_annot$probe){
hgnc_annot$Gene.Symbol[hgnc_annot$probe == "CDCA1"] = NA
}
# Final steps for annotation data frames
full_dups_filt = full_dups_filt %>% dplyr::select(probe, Entrez) %>%
dplyr::rename(EntrezGene.ID=Entrez)
Join_final = unique(rbind(Join_pre, full_dups_filt))
Join_annot = as.data.frame(expressionDF[,1])
colnames(Join_annot) = "probe"
Join_annot2 = Join_annot %>% left_join(Join_final)
Join_annot2$EntrezGene.ID[Join_annot2$probe %in% final_check$probe] = NA
Join_annot2$EntrezGene.ID[Join_annot2$probe %in% dup_names] = NA
dannot = distinct(Join_annot2)
# Do a PAM50 check
data("pam50")
inconsistencies = as.data.frame(pam50$centroids.map[,c("probe", "EntrezGene.ID")])
inconsistencies[51,] = c("AURKA", "6790")
rownames(inconsistencies) = as.character(inconsistencies$probe)
dannot$f_probe = as.character(dannot$probe)
for (j in 1:length(inconsistencies$EntrezGene.ID)){
dannot$f_probe[dannot$EntrezGene.ID == inconsistencies$EntrezGene.ID[j]] = inconsistencies$probe[j]
}
if ("CDCA1" %in% dannot$probe &&
"NUF2" %in% dannot$probe){
dannot$EntrezGene.ID[dannot$probe == "CDCA1"] = NA
}
sigs = length(intersect(dannot$EntrezGene.ID, pam50$centroids.map$EntrezGene.ID))
message(paste0(sigs, " pam50 signatures were found"))
dannot$EntrezGene.ID[dannot$probe %in% final_check$probe] = NA
dannot$EntrezGene.ID[dannot$probe %in% dup_names] = NA
expressionDF = expressionDF %>% inner_join(dannot, by = "probe")
# If after all this process there still are duplicate probes
# keep the most variant probe for this gene
last_dups = dannot$f_probe[which(duplicated(dannot$f_probe) == TRUE)]
last_dups_probe = dannot$probe[dannot$f_probe %in% unique(last_dups)]
if (length(last_dups) != 0){
means = expressionDF[expressionDF$probe %in% last_dups_probe,]
means = means %>% dplyr::select(-probe, -EntrezGene.ID) %>% group_by(f_probe) %>%
summarise_all(mean, na.rm = TRUE)
for(i in 1:nrow(expressionDF)){
if(expressionDF$f_probe[i] %in% means$f_probe){
expressionDF[i,2:(ncol(expressionDF)-2)] = means[means$f_probe==expressionDF$f_probe[i],
2:ncol(means)]
}
}
expressionDF = expressionDF %>%
dplyr::select(-f_probe, -EntrezGene.ID) %>% inner_join(dannot)
if ("CDCA1" %in% dannot$probe &&
"NUF2" %in% dannot$probe){
expressionDF$f_probe[expressionDF$probe == "CDCA1"] = "CDCA_ALIAS"
}
expressionDF = expressionDF %>%
dplyr::select(-probe, -EntrezGene.ID) %>%
dplyr::rename(probe=f_probe) %>%
dplyr::select(probe, everything())
} else {
expressionDF = expressionDF %>%
dplyr::select(-probe, -EntrezGene.ID) %>%
dplyr::rename(probe=f_probe) %>%
dplyr::select(probe, everything())
}
if ("CDCA1" %in% dannot$probe &&
"NUF2" %in% dannot$probe){
dannot$f_probe[dannot$probe == "CDCA1"] = "CDCA_ALIAS"
}
dannot = dannot %>% dplyr::select(-probe) %>%
dplyr::rename(probe=f_probe) %>% distinct()
rownames(dannot) = as.character(dannot[,"probe"])
message("Annotation data frame created successfully")
Entrez = expressionDF %>% inner_join(dannot) %>%
dplyr::select(-probe) %>%
dplyr::select(EntrezGene.ID, everything())
HGNC = expressionDF %>% inner_join(hgnc_annot) %>%
dplyr::select(-probe) %>%
dplyr::select(Gene.Symbol, everything())
return(list(Entrez = as.data.frame(Entrez), HGNC = as.data.frame(HGNC),
dannot_Entrez = as.data.frame(dannot),
dannot_HGNC = as.data.frame(hgnc_annot)))
}
# The Genefu predictions function - full #####
Genefu_predictions <- function(expressionDF, HGNC, dannot_Entrez, dannot_HGNC){
library(genefu)
library(dplyr)
library(org.Hs.eg.db)
# The purpose of this function is to generate genefu predictions
# based on properly Entrez-annotated datasets. expressionDF is the Entrez
# exprs and HGNC is the HGNC exprs for intClust
entrezgenes = as.character(expressionDF$EntrezGene.ID)
hgncgenes = as.character(HGNC$Gene.Symbol)
expressionDF2 <- expressionDF %>% dplyr::select(-EntrezGene.ID)
HGNC2 <- HGNC %>% dplyr::select(-Gene.Symbol)
exprsDF <- t(expressionDF2)
exprsHGNC <- t(HGNC2)
colnames(exprsDF) = entrezgenes
colnames(exprsHGNC) = hgncgenes
# Predictions
# Mammaprint
# std = "robust": Standardization based on the 0.025 and 0.975 quantiles
# same as in the case of
data("sig.gene70")
RNGversion("4.0.2")
set.seed(123)
mammaprint <- gene70(data = exprsDF, annot = dannot_Entrez, do.mapping = TRUE,
std = "robust")
mammaprint_score <- mammaprint$score %>% as.data.frame()
mammaprint_risk <- mammaprint$risk %>% as.data.frame()
message("Mammaprint: OK")
# rorS
RNGversion("4.0.2")
set.seed(123)
RORS <- rorS(data = exprsDF, annot = dannot_Entrez, do.mapping = TRUE)
rorS_score <- RORS$score %>% as.data.frame()
rorS_risk <- RORS$risk %>% as.data.frame()
message("rorS: OK")
# EndoPredict
# This algorithm performs the following rescaling process for values to
# approximate the scale of the original Affymetrix data:
# https://rdrr.io/bioc/genefu/src/R/endoPredict.R
data("sig.endoPredict")
RNGversion("4.0.2")
set.seed(123)
EndoPredict = endoPredict(data = exprsDF, annot = dannot_Entrez, do.mapping = TRUE)
endoPredict_score <- EndoPredict$score %>% as.data.frame()
endoPredict_risk <- EndoPredict$risk %>% as.data.frame()
message("EndoPredict: OK")
# Oncotype DX
data("sig.oncotypedx")
RNGversion("4.0.2")
set.seed(123)
OncotypeDX = oncotypedx(data = exprsDF, annot = dannot_Entrez, do.mapping = TRUE,
do.scaling = TRUE)
oncotypedx_score <- OncotypeDX$score %>% as.data.frame()
oncotypedx_risk <- OncotypeDX$risk %>% as.data.frame()
message("Oncotype DX: OK")
# When choosing ".robust" versions of the algorithm, scaling of the expression
# values is performed using the rescale() function of the genefu package. This function
# rescales values x based on quantiles specified by the user such that
# x’ = (x - q1) / (q2 - q1) where q is the specified quantile, q1 = q / 2,
# q2 = 1 - q/2) and x’ are the new rescaled values. This version of scaling is
# more concordant with clinical experience. The robust version of the algorithm
# is by default chosen when these algorithms are run.
# SCMGENE
data("scmgene.robust")
scmgene <- molecular.subtyping(sbt.model = "scmgene", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("scmgene: OK")
# SCMOD1
data("scmod1.robust")
scmod1 <- molecular.subtyping(sbt.model = "scmod1", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("scmod1: OK")
# SCMOD2
data("scmod2.robust")
scmod2 <- molecular.subtyping(sbt.model = "scmod2", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("scmod2: OK")
# PAM50
data("pam50.robust")
PAM50 <- molecular.subtyping(sbt.model = "pam50", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("pam50: OK")
# ClaudinLow
# This algorithm by default performs median centering
data("claudinLowData")
ClaudinLow <- molecular.subtyping(sbt.model = "claudinLow", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("ClaudinLow: OK")
# SSP2006
data("ssp2006.robust")
ssp_2006 <- molecular.subtyping(sbt.model = "ssp2006", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("ssp2006: OK")
# SSP2003
data("ssp2003.robust")
ssp_2003 <- molecular.subtyping(sbt.model = "ssp2003", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("ssp2003: OK")
# GENIUS
# rescaling here is performed to yield more robust comparisons with AURKA
# ESR1 and ERBB2 modules
data("sig.genius")
RNGversion("4.0.2")
set.seed(123)
GENIUS_score <- genius(data = exprsDF, annot = dannot_Entrez,
do.mapping = TRUE, do.scale = TRUE)$score %>%
as.data.frame()
message("GENIUS: OK")
# GGI - similar logic to the Nottingham Index score
data("sig.ggi")
RNGversion("4.0.2")
set.seed(123)
GGI <- ggi(data = exprsDF, annot = dannot_Entrez, do.mapping = TRUE)
GGI_score <- GGI$score %>% as.data.frame()
GGI_risk <- GGI$risk %>% as.data.frame()
message("GGI: OK")
# PIK3CA_GS
data("sig.pik3cags")
RNGversion("4.0.2")
set.seed(123)
PIK3CA_GS <- pik3cags(data = exprsDF, annot = dannot_Entrez, do.mapping = TRUE) %>%
as.data.frame()
message("PIK3CA_GS: OK")
# IC10
# This algorithm normalises features by default (z-score)
intClust <- molecular.subtyping(sbt.model = "intClust", data = exprsHGNC,
annot = dannot_HGNC, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("iC10 OK")
# Finalising
predictions <- cbind(scmgene, scmod1, scmod2, PAM50, ClaudinLow, ssp_2006,
ssp_2003, GENIUS_score, GGI_score, GGI_risk, PIK3CA_GS,
mammaprint_risk, mammaprint_score, rorS_risk, rorS_score,
endoPredict_risk, endoPredict_score, oncotypedx_risk,
oncotypedx_score, intClust)
colnames(predictions) <- c("scmgene", "scmod1", "scmod2", "pam50", "ClaudinLow",
"ssp2006", "ssp2003", "GENIUS_score", "GGI_score",
"GGI_risk", "PIK3CA_GS", "Mammaprint_risk",
"Mammaprint_score", "rorS_risk", "rorS_score",
"EndoPredict_risk", "EndoPredict_score",
"OncotypeDX_risk", "OncotypeDX_score", "IC10")
predictions$Sample.ID = as.character(rownames(predictions))
predictions <- predictions %>% dplyr::select(Sample.ID, everything())
return(predictions)
}
# Feed one expression matrix.
# The Genefu predictions function - fewer predictors #####
Genefu_predictions_2 <- function(expressionDF, HGNC, dannot_Entrez, dannot_HGNC){
library(genefu)
library(dplyr)
library(org.Hs.eg.db)
# The purpose of this function is to generate genefu predictions
# based on properly Entrez-annotated datasets. expressionDF is the Entrez
# exprs and HGNC is the HGNC exprs for intClust
entrezgenes = as.character(expressionDF$EntrezGene.ID)
hgncgenes = as.character(HGNC$Gene.Symbol)
expressionDF2 <- expressionDF %>% dplyr::select(-EntrezGene.ID)
HGNC2 <- HGNC %>% dplyr::select(-Gene.Symbol)
exprsDF <- t(expressionDF2)
exprsHGNC <- t(HGNC2)
colnames(exprsDF) = entrezgenes
colnames(exprsHGNC) = hgncgenes
# Predictions
# Mammaprint
# std = "robust": Standardization based on the 0.025 and 0.975 quantiles
# same as in the case of
data("sig.gene70")
RNGversion("4.0.2")
set.seed(123)
mammaprint <- gene70(data = exprsDF, annot = dannot_Entrez, do.mapping = TRUE,
std = "robust")
mammaprint_score <- mammaprint$score %>% as.data.frame()
mammaprint_risk <- mammaprint$risk %>% as.data.frame()
message("Mammaprint: OK")
# rorS
RNGversion("4.0.2")
set.seed(123)
RORS <- rorS(data = exprsDF, annot = dannot_Entrez, do.mapping = TRUE)
rorS_score <- RORS$score %>% as.data.frame()
rorS_risk <- RORS$risk %>% as.data.frame()
message("rorS: OK")
# When choosing ".robust" versions of the algorithm, scaling of the expression
# values is performed using the rescale() function of the genefu package. This function
# rescales values x based on quantiles specified by the user such that
# x’ = (x - q1) / (q2 - q1) where q is the specified quantile, q1 = q / 2,
# q2 = 1 - q/2) and x’ are the new rescaled values. This version of scaling is
# more concordant with clinical experience. The robust version of the algorithm
# is by default chosen when these algorithms are run.
# SCMOD1
data("scmod1.robust")
scmod1 <- molecular.subtyping(sbt.model = "scmod1", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("scmod1: OK")
# PAM50
data("pam50.robust")
PAM50 <- molecular.subtyping(sbt.model = "pam50", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("pam50: OK")
# ClaudinLow
# This algorithm by default performs median centering
data("claudinLowData")
ClaudinLow <- molecular.subtyping(sbt.model = "claudinLow", data = exprsDF,
annot = dannot_Entrez, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("ClaudinLow: OK")
# IC10
# This algorithm normalises features by default (z-score)
intClust <- molecular.subtyping(sbt.model = "intClust", data = exprsHGNC,
annot = dannot_HGNC, do.mapping = TRUE)$subtype %>%
as.data.frame()
message("iC10 OK")
# Finalising
predictions <- cbind(scmod1, PAM50, ClaudinLow, mammaprint_risk, mammaprint_score,
rorS_risk, rorS_score, intClust)
colnames(predictions) <- c("scmod1", "pam50", "ClaudinLow", "Mammaprint_risk",
"Mammaprint_score", "rorS_risk", "rorS_score", "IC10")
predictions$Sample.ID = as.character(rownames(predictions))
predictions <- predictions %>% dplyr::select(Sample.ID, everything())
return(predictions)
}
# Feed one expression matrix.
