HairpinSVM and DroshaSVM. (A) The flow chart of HairpinSVM: for each pre-miRNA in miRBase, it is first mapped to the genome of Drosophila melanogaster and extended to 180nt. These 180nt sequences are collected into the Sample DB. For each sample in the Sample DB, all of its subsequences longer than 50nt are folded by RNAfold. After elimination of redundancy folding results, the ones same with the structure in miRBase are assembled in the true training set, other hairpin structures are assembled in false training set. (B) The flow chart of DroshaSVM: for each pre-miRNA hairpin structure, all the sites of 5' arm are accepted as Drosha processing site candidate, the true processing sites are based on miRBase annotation. Other sites compose false training set.