Large RNA sequences ? Does it has any sense ?

[perencia]

Hi!

First, i'm a computer scientist recently exploring bioinformatics field, so please forgive me if i say something really stupid

Basically i'm studying the possibility of implementing Nussinov-Jacobsen algorithm on GPU's, accelerating, if possible, time performance in orders of magnitude; but to accomplish that, the RNA sequence has to be very large. I was wondering if it has some sense since i've seen most RNA seqs are about 200 bases.

Thanks!

[raela]

How long do you mean by 'very large'? It depends on the sequencing technology used and the length ordered. Even 200 is somewhat in the 'long' range for NGS (I believe).

[epigen]

What kind of RNA do you mean? There are entire RNA genomes of bacteria. And normal mRNAs are some 100s to 1000s nucleotides long.
By quickly googling Nussinov-Jacobsen I learned that you can do RNA folding prediction with it. That only makes sense for small RNAs.

[krobison]

454 has 400+ base reads & PacBio is promising reads that long or much longer.

Folding of longer RNAs could be interesting, as secondary structure is sometimes involved in the stability, localization or utilization of an RNA.

It's a niche, but that doesn't mean it isn't interesting.

[mrawlins]

Many RNAs are long, but the current sequencing technologies fragment them prior to sequencing, since they perform better on shorter sequences. SOLiD works up to 50 bp, Illumina works up to about 100 bp, and 454 can get a few hundred bp. If you want something larger you'll have to piece together multiple reads into a longer consensus sequence.
It seems to me, though, that if you need a longer consensus sequence you could just use the complement of the genomic sequence (which is the RNA sequence) for some interesting genes. If your goal is to demonstrate an algorithmic speedup using a GPU-based approach it seems that it wouldn't be important to have cutting-edge RNA data, but it would be better to use a well studied RNA (like ribosomal RNA or tRNA) for your comparison.

[perencia]

Many RNAs are long, but the current sequencing technologies fragment them prior to sequencing, since they perform better on shorter sequences. SOLiD works up to 50 bp, Illumina works up to about 100 bp, and 454 can get a few hundred bp. If you want something larger you'll have to piece together multiple reads into a longer consensus sequence.
It seems to me, though, that if you need a longer consensus sequence you could just use the complement of the genomic sequence (which is the RNA sequence) for some interesting genes. If your goal is to demonstrate an algorithmic speedup using a GPU-based approach it seems that it wouldn't be important to have cutting-edge RNA data, but it would be better to use a well studied RNA (like ribosomal RNA or tRNA) for your comparison.

Ok.

And how many nucleotides can have that ribosomal or tRNA ?

[mrawlins]

In Shewanella the longest ribosomal sequence is about 2900 bases long. Human ribosome sequences may be a bit longer. The tRNAs (in Shewanella) are about 76 bases long. I seem to recall that tRNAs have some of the best documented secondary structure, though, so while they may not make a good test of your algorithm's speed, they might be a good test of accuracy.

[perencia]

In Shewanella the longest ribosomal sequence is about 2900 bases long. Human ribosome sequences may be a bit longer. The tRNAs (in Shewanella) are about 76 bases long. I seem to recall that tRNAs have some of the best documented secondary structure, though, so while they may not make a good test of your algorithm's speed, they might be a good test of accuracy.

Thanks!

I wonder, what about the searching for structures on an entire genome ?

[maubp]

I wonder, what about the searching for structures on an entire genome ?

Only some viruses have an RNA genome (most organisms use DNA), and their genomes tend not to be very big (not big enough to worry about GPU optimisations I would guess).

[Bruins]

Have you had a chance to take a look at Rfam and Sean Eddy's Infernal?

[perencia]

Have you had a chance to take a look at Rfam and Sean Eddy's Infernal?

No, but i'll look for them now

I've searching a little more, and found that report

http://hal.archives-ouvertes.fr/docs/00/42/55/43/PDF/gpufold_ICCS.pdf

and that implementation

http://www.cc.gatech.edu/~bader/papers/GTfold.html

Former is a GPU implementation of the Unafold Algorithm
( http://mfold.bioinfo.rpi.edu/ )

It seems that a GPU optimisation may take room on a multiple RNA structure prediction, as in the first report ( a set of 11 Picor-
naviral sequences (7124 to 8214 nucleotides)).
I'll post what i find

Anyway, i'd like to known which are the benefits from such procedures, where do they impact. Bioinformatics is a large field i guess .