So ... I am working on a revised version of CleaveLand2 to fix these issues and hopefully improve performance. The major philosophical difference between CleaveLand2 and the CleaveLand 1 version that we published on is the use (in CleaveLand2) of extensive randomized target predictions to estimate an ad-hoc p-value. I really like this approach, but it is computationally expensive to get alignments of 1,000 random miRNA queries for each real miRNA query. In my revision, I may try and find a shortcut around this ....
So, anyway, for people who are waiting for me to fix these issues, please stay tuned. I'm on it. -- mike axtell