Aligning mRNA against Riken database

[cswarth]

I am volunteering in a lab and have been tasked with analysis of some sequences of short mRNA fragments. In a previous life I was a programmer but I have no experience in bioinformatics so this is all new to me.

The plan is to use bowtie to align the cDNA fragments to the Riken mouse cDNA database. The problem is that when I look at that database (ftp://fantom.gsc.riken.jp/fantomdb/3.0/) it looks quite different from mRNA fragments that I download from the UCSC browser.

From the UCSC browser I can get sequences plus a table that has txStart/txEnd and cdsStart/cdsEnd, so I know which chromosome and where on the chromosome each fragment maps, and I know how long the 5' and 3' UTRs are. The Riken database doesn't seem to have that information. It occasionally has some cds location info, but generally not, and nothing about where the fragment maps onto the chromosome.

So my questions are, am I missing something about the Riken data that would make it more useful? If not I will try to assemble a set of non-overlapping sequences from the UCSC browser.

Second, is this a reasonable approach to analyzing short-reads from RNA-seq? Would it be better to use something like cufflinks to align against a full mouse genome?

Thanks in advance for any help.

[john_mu]

Looking at that data, the reads don't look very short... They don't look like they are from a high-throughput sequencer.

For reads that long (and since you don't have many reads), your best best is probably to align them with BLAT. They should align pretty confidently.

EDIT: actually, sorry, they are already annotated... I'm not too familiar with this, you'll have to wait for someone else.

[cswarth]

Originally Posted by john_mu

For reads that long (and since you don't have many reads), your best best is probably to align them with BLAT. They should align pretty confidently.

Your looking at the Riken database? Those aren't our reads, those are what I want to align our reads to.

We will be getting about 25 millions reads from each sample, and there will be many samples. blat isn't an option for that many, right? In any case, I am searching for the right thing to align mouse cDNA reads against that at least has coding regions annotated, and if possible also tissue source and protein product annotations. I don't know if that is a reasonable expectation.

[john_mu]

oh... so you want to align your short reads against only the known annotated coding regions?

I don't think that is necessary though. If you align them to the entire genome with TopHat or SpliceMap, 90%+ of the reads will align to the coding regions anyway. RNA-seq is quite specific.

After the alignment, then you can compare your results with known annotations and try to see what happened.

Is there any particular reason, you want to only align the reads to known coding regions?

[cswarth]

Originally Posted by john_mu

Is there any particular reason, you want to only align the reads to known coding regions?

Frankly I am doing it that way because that is what I was told to do by the PI. It makes sense if the Riken database is clean and has annotations that lead back to well known gene names and protein products. And the details of our experimental setup really guarantees we will only see reads in coding regions.

But the more I look at the Riken database, the less I like this approach. I'll install cufflink and tophat and see how those can help me align against the whole genome.

Thank you for the reply.

[john_mu]

oh I see.. well hope it works out! Good luck