CMfinder—A Covariance Model Based RNA Motif Finding Algorithm: Appendix, additional technical details about the algorithm

We used RNAfold in the Vienna package [1] to compute the minimal free energy for all subsequences in a given sequence. The ones whose length and number of stem-loops are within the range (default 30 ∼ 100 bases, 1 ∼ 2 stem loops), and are locally optimal (base paired at the ends, with no lower-energy states by extending or shrinking 2 bases at the ends) are selected. They are then sorted by the energy scaled by sequence length, and candidates are selected from the top of the list. We allow overlapped candidates as long as one of them is significantly longer than the other one.