An example of a reverse-complementary distance-sensitive n-gram profile (RCDSNGP) representation with n = 4, 5, and 6 for a given sequence (AAGCTTGAGACACAGCT) with the reference subsequence marked in bold*.

*Given an m-length sequence s = s1, s2… si…si+j…sm, the RCDSNGP of s with respect to an j-length reference subsequence x = si…si+j−1 is a set of K 2-tuples, denoted as RCDSNGP(s) RCDSNGP(s) = {({f1, r1, d1}, c1),({f2, r2, d2}, c2)…({fK, rK, dK}, cK)}, fk being a distinct n-gram, rk being the reverse complement of fk, dk being the relative distance parameter, and ck being the sum of frequency counts of fk and rk with the same dk relative to x in s. Each set in a 2-tuple ({fk, rk, dk}) is a reverse-complementary distance-sensitive n-gram (RCDSNG), or a feature in our study. This RCDSNGP representation was adopted for all training sequences. For testing processes, each sequence was converted to RCDSNGP first, and then represented according to the selected RCDSNGs generated from the training datasets, including those with zero count.