e1071 package, svm method

[Dalene]

Hi, I want to use the e1071 package to do the following:

I have a peptidomics dataset of 890 peptides. In this cohort, there are normotensive and hypertensive individuals. I want to use support vector machines to identify peptide biomarkers that differentiates between normotensive and hypertensive individuals.

I understand from the other examples that you can get the target variable (for example in the iris dataset, you can get the "species" using svm method), however I am not sure how to identify more than one biomarker based on blood pressure? I will also have a training and testing group with equal amounts of normotensive and hypertensive individuals. Therefore, I want to identify peptide biomarkers in the training group and test the peptide biomarkers in the testing group. Any help?

[Max]

How many samples do you have?

You might want a different model. SVMs don't really help with feature selection (i.e. identifying important peptides in you data). If you want to fit a logistic model, you would be better off using the lasso technique (via the glmnet package) or, if you want a nonlinear model, try a tree-based models such as rpart.

If you have to use SVM's consider them in concert with recursive feature elimination, perhaps using caret. I do think the other models are a better choice though.