I have a dataframe with genes and samples (cancer vs. normal) and I already did LASSO and cross validation to choose best lambda, as well as finding the genes with non-zero coefficients (x in code below is my dataframe containing these). What I want to do next is add another column to x which contains the gene symbols (symbol from original dataframe daf) corresponding to those gene with non-zero coefficients in x. I have been trying for over an hour now to get that to work but haven't had success. Any suggestions on what is the best way to do that? Below is my code:
probeID<-c("213456_at", "217428_s_at", "219230_at", "226228_at","230030_at")
symbol<-c("SOSTDC1","COL10A1", "TMEM100", "AQP4", "HS6ST2")
BCR1<-c(28.005966, 30.806433, 17.341375, 17.40666, 30.039436)
BCR2<-c(30.973469, 29.236025, 30.41161, 20.914383, 20.904331)
BCR3<-c(26.322796, 25.542833, 22.460772, 19.972183, 30.409641)
BCR4<-c(26.441898, 25.837685, 23.158352, 20.379173, 33.81327)
BCR5<-c(39.750206, 19.901133, 28.180124, 22.668673, 25.748884)
CTL6<-c(23.004385, 28.472675, 23.81621, 26.433413, 28.851719)
CTL7<-c(22.239546, 28.741674, 23.754929, 26.015385, 28.16368)
CTL8<-c(29.590443, 30.041988, 21.323061, 24.272501, 18.099016)
CTL9<-c(15.856442, 22.64224, 29.629637, 25.374926, 22.356894)
CTL10<-c(38.137985, 24.753338, 26.986668, 24.578161, 19.223558)
daf<-data.frame(probeID, symbol, BCR1, BCR2, BCR3, BCR4, BCR5, CTL6, CTL7, CTL8, CTL9, CTL10)
daf1<-t(daf[,3:12])
colnames(daf1)<-daf$probeID
View(daf1)
Type<-c("cancer", "cancer", "cancer", "cancer", "cancer", "normal", "normal", "normal", "normal", "normal")
Sample<-c("BCR1", "BCR2", "BCR3", "BCR4", "BCR5", "CTL6", "CTL7", "CTL8", "CTL9", "CTL10")
type.df<-data.frame(Sample, Type)
daf2<-data.frame(daf1, type.df$Type)
names(daf2)[names(daf2) == "type.df.Type"] <- "Type"
View(daf2)
daf2$Type<-as.factor(daf2$Type)
lassoModel <- glmnet(
x=data.matrix(daf2[,-6]),
y=daf2$Type,
alpha=1,
family="binomial")
plot(lassoModel, xvar="lambda")
coef(lassoModel)[,5][coef(lassoModel)[,5]!=0]
#Cross Validation
cv.lassoModel<- cv.glmnet(
x=data.matrix(daf2[,-6]),
y=daf2$Type,
alpha=1, family="binomial")
# plot variable deviances vs. shrinkage parameter, λ (lambda)
plot(cv.lassoModel)
#Chose best lambda
idealLambda <- cv.lassoModel$lambda.min
idealLambda1se <- cv.lassoModel$lambda.1se
print(idealLambda); print(idealLambda1se)
# derive coefficients for each gene
co <- coef(cv.lassoModel, s=idealLambda, exact=TRUE)
co
co.se <- coef(cv.lassoModel, s=idealLambda1se, exact=TRUE)
co.se
#Select those genes that have non-zero coefficients for the best lambda
cv.glm.probe<-coef(cv.lassoModel, s="lambda.min")
x<-data.frame(cv.glm.probe[cv.glm.probe[,1]!=0,])
If you look at your coefficients, they have an extra "X" in front, because glm, lm, glmnet etc does not like variables that start with a number, and by default adds an "X".
x$symbol = daf$symbol[match(sub("^X","",rownames(x)),daf$probeID)]
x
cv.glm.probe.cv.glm.probe...1.....0... symbol
(Intercept) -41.23471919 <NA>
X217428_s_at 0.18134947 COL10A1
X226228_at 1.61933359 AQP4
X230030_at -0.03797544 HS6ST2
If you have say a data.frame where the X doesn't exist, for example:
df = data.frame(coefficients=runif(5))
rownames(df) = sample(daf$probeID,5)
df$symbol = daf$symbol[match(rownames(df),daf$probeID)]
Or have the actual probe in a data frame and merge:
df = data.frame(probe=sample(daf$probeID,5),coefficients=runif(5))
merge(df,daf[,c("probeID","symbol")],by.x="probe",by.y="probeID")
probe coefficients symbol
1 213456_at 0.40697051 SOSTDC1
2 217428_s_at 0.97655456 COL10A1
3 219230_at 0.09496977 TMEM100
4 226228_at 0.70865375 AQP4
5 230030_at 0.35125967 HS6ST2