short reads clustering

[capricy]

I have 20 million short reads: 25 bp each, and there is no genome reference available.

I would like to examine the patterns of those reads, that is, some reads should be derived from the same gene locus.

Any suggestions about a quick method to cluster them together so that I know how many loci they might be derived from?

Thank you very much for any thoughts?

[Brian Bushnell]

I think you're asking for the impossible. Even if 25bp reads cluster together, that's not an indication they came from the same place. Clustering algorithms are pretty bad even with 150bp reads from different species, and clustering is notoriously bad at generating the correct number of clusters.

I suggest mapping them to the reference of the most closely related organism you can find, or attempting assembly with K=21 with, say, Velvet. Then mapping them to the contigs you get, and blasting the contigs to see what they are. But it might be more cost-effective to sequence longer reads (150bp - 250bp) and make a good assembly than to waste your time on 25bp reads with no reference.