randomForest: how to choose predictor variables based on multiclass response variable?

Question

[shripathiacademics]

Hi, I am trying to use randomForest (classification) in R to get list of SNP markers which will explain most of the population genetic structure in the data set. I tried to do a test run, and obtained importance score per class factor (means per population), mean decrease in accuracy and mean decrease in Gini. My quesiton is: How to choose top SNPs when there are multiple importance scores?

Thanks in advance. SA