Classification performance.

The percentages are two-fold cross validation accuracies for the classification of LNFRs showing remodeling versus randomly selected non-LNFR DNA sequences from the human genome. Altogether four sets of remodeled LNFRs were considered: GC-rich and AT-rich remodeled LNFRs, both for resting and activated T cells. Each row corresponds to one choice of , the sub-sequence length parameter of the spectrum kernel. GC-rich remodeling LNFRs (third and fifth column) can be classified with higher accuracy than AT-rich remodeling LNFRs (second and fourth column). The highest accuracy for GC-rich remodeling LNFRs in resting T cells is 68.6% and is achieved for .