having various length of upstream and downstream region (0, 25, 50, 75, 100bp).
The procedure of cross-validation is the same as described in the main text.
The results were shown in Figure S1.
Figure S1. Sensitivity and specificity in test data. '0
bp', '25 bp', '50
bp', '75 bp'
and '100 bp':
performance of HMMs trained using training samples with 0, 25, 50,
75 and 100 bp upstream and downstream region, respectively.
The performance of HMMs trained using training samples without surrounding
region (0 bp) is the worst. Specificity of '75 bp' and '100 bp' is higher when
sensitivity is < 0.5. However, these HMMs missed detecting miRNAs when two
miRNAs are closely located. This is because we used the viterbi decoding algorithm
to detect miRNAs. The viterbi decoding can not produce overlapping predictions.
Therefore, using 25 or 50 bp upstream and downstream regions is suitable for
our method.