Gene Selection with SVM-RFE in R

This example demonstrates how to use Support Vector Machine Recursive Feature Elimination (SVM-RFE) in R to identify the most relevant genes influencing a result variable. We assume a dataframe named 'dataframe' containing 12 gene symbols as variables (columns 1-12) and a 'result_var' variable (column 13).

R Code

# Load required packages
library(e1071)
library(caret)

# Define the SVM model
svm_model <- svm(result_var ~ ., data = dataframe, kernel = 'linear')

# Perform SVM-RFE
svm_rfe <- rfe(dataframe[,1:12], dataframe$result_var, sizes = c(1:12), rfeControl = rfeControl(functions = svmFuncs, method = 'cv', number = 10))

# Print the ranking of genes
print(svm_rfe$optVariables)

# Plot the SVM-RFE results
plot(svm_rfe, type=c('g', 'o'))

Explanation

1. Load Packages: We load the e1071 package for SVM functionality and the caret package for SVM-RFE implementation.
2. SVM Model: We define a linear kernel SVM model using all variables except the result variable (result_var ~ .).
3. SVM-RFE: The rfe function performs SVM-RFE. We provide the gene variables (columns 1-12), the result variable (dataframe$result_var), the range of feature sizes to consider (1 to 12), and control parameters for 10-fold cross-validation (rfeControl).
4. Print Gene Ranking: The optVariables attribute of the svm_rfe object holds the ranked genes based on relevance to the result variable.
5. Plot Results: The plot function visualizes the SVM-RFE results, showing model accuracy across different feature sizes. The 'g' and 'o' arguments generate both a line and scatter plot for clarity.

This code will identify the most relevant genes impacting the result variable using the powerful SVM-RFE approach. You can adjust the sizes parameter and rfeControl parameters for more fine-grained control and optimization.