Map > Problem Definition > Data Preparation > Data Exploration > Modeling > Classification
 

Modeling - Classification
I. Data Preparation

1- Load libraries
library(data.table)
library(formattable)
library(plotrix)
library(limma)
library(dplyr)
library(Rtsne)
library(MASS)
library(xgboost)
 

2- Read Expressions file
df <- read.csv("GSE74763_rawlog_expr.csv")
df2 <- df[,-1]
rownames(df2) <- df[,1]
expr <- transpose(df2)
rownames(expr) <- colnames(df2)
colnames(expr) <- rownames(df2)
dim(expr)
 

3- Read Samples file
targets <- read.csv("GSE74763_rawlog_targets.csv")
colnames(targets)
dim(targets)
 

4- Merge Expressions with Samples
data <- cbind(expr, targets)
colnames(data)
dim(data)
 
II. Variable Selection
Here, we use t-test to select the topN variables (probes) from 9480 probes.
d1 <- data

groups <- as.fumeric(data$target)
tval <- lapply(d1, function(x) { t.test(x ~ groups)$statistic })

tdf <- as.data.frame(t(as.data.frame(tval)))
colnames(tdf)
tdf$t <- abs(tdf$t)
tdfo <- tdf[order(tdf$t,decreasing=TRUE),,drop=FALSE]
head(tdfo)

III. Logistic Regression
d1 <- data
d1$geo_accession <- NULL
d1$age <- NULL
d1$gender <- NULL

d1$target <- as.fumeric(d1$target)-1
model <- glm(target ~ P000833+P007414+P002449, data = d1, family = binomial, maxit = 100)
print(model)

 
IV. XGBoost (Extreme Gradient Boosting)
XGBoost does not need a variable selection step.
d1 <- data
d1$geo_accession <- NULL
d1$age <- NULL
d1$gender <- NULL

d1$target <- as.fumeric(d1$target)-1
bst <- xgboost(data = as.matrix(d1), label = d1$target, max.depth = 5, eta = 1, nthread = 2, nrounds = 100
, objective = "binary:logistic")

var_importance <- xgb.importance(model = bst)
print(var_importance)

 
V. Bioada SmartArray
Watch this video to learn how you can build classification models using Bioada SmartArray significantly faster and easier.