Choose a read dataset for the paper

blahah / assemblotron-paper

Paper for Assemblotron

MIT License

0 stars 0 forks source link

Choose a read dataset for the paper #7

Open blahah opened 9 years ago

blahah commented 9 years ago

Features of the dataset:

from a species with a decent reference genome that is well annotated (or at least, as well annotated as any reference, which is not very well)
fairly small dataset, or with a sample or replicate that is fairly small, so it can be used for full parameter sweep without subsetting
when assembled, should not knot up the assembler (this will hugely slow down the full sweep)

blahah commented 9 years ago

This yeast dataset, generated for the Trinity paper, is an option:

full study dataset: http://www.ebi.ac.uk/ena/data/view/SRP005611
we would use a single run from the study, e.g. this one which has 3.7 million read pairs (file size 215M per file)

Advantages of this dataset include:

it's yeast, so the transcriptome is fairly small
there is still a considerable amount of splicing
we can just cite the paper it came from (true for almost anything in ENA)

blahah commented 9 years ago

Doing some preliminary analysis on the yeast dataset, I downloaded it and ran assemblotron without subsampling. A single assembly with SOAPdenovoTrans + transrate takes about 4 minutes on 24 cores of our cluster, so this looks promising for a sweep of perhaps 4 major parameters (K, d, e, t). Unfortunately there's some weird combination of the environment on the cluster and the build of Salmon that is making it segfault on some data. I am working through this with Rob Patro (Salmon developer).

blahah commented 9 years ago

Turns out there was a bug in Salmon that was triggered by this dataset. It's now fixed in Salmon 0.4.2 and transrate 1.0.0, so I've set the parameter sweep running again.

blahah commented 9 years ago

Possible arabidopsis read datasets:

https://www.ebi.ac.uk/ena/data/view/DRX014494