Clustering Of Ncrna

Question

[cacaucenturion]

I want to cluster a set of ncRNAs in human into different families based on sequence similarity. How should I do it and which software might be suitable to to it? Thanks very much!

Answer 1

[Ryan Thompson]

You might want to map your ncRNAs to the genome and then cluster based on overlap, using something like blockbuster.

[cacaucenturion]

Thanks! Actually what I meant was that I wanted to cluster RNA into families. Sorry for not saying clearly.

Answer 2

[Paul Gardner]

I'd build HMMs/CMs from alignments of the RNAs. Then iteratively add similar sequences to the largest clusters until you're done.

E.g. run hmmbuild -> hmmsearch -> hmmalign -> (predict a secondary structure w/ e.g. RNAalifold) -> make a stockholm alignment w/ secondary structure annotation -> cmbuild -> cmsearch -> cmalign. Repeat last 3 stages until convergence. Repeat with unaligned sequences.

[cacaucenturion]

Thanks very much! So if I have a lot of RNA sequences and do not know which families they should belong to, how can I build HMM? Thanks!

[cacaucenturion]

Actually my question is that what sequences I should choose to start making alignments and building HMM? Can I use another kind of software like CDhit to get a clustering result initially and then use the largest cluster to build the HMM? Thanks!

Answer 3

[Neilfws]

In general, the answer to "how do I cluster sequences?" is CD-HIT.

In this case, specifically CD-HIT-EST. From the applications page:

CD-HIT-EST has been used in clustering many types of sequences such as Expressed Sequence Tags (ESTs), MicroRNAs (miRNAs) (RNA, 2007 13:170-187), oligonucleotide probes (Bioinformatics, 2007 23:1195), 16S rRNA sequences (Nature, 2009, 457:480).