SOLiD Accuracy Enhancement Tool (SAET) multicore and ref genome size questions

[KevinLam]

Just dl SAET to try

one of the options jumped out at me

refLength – expected length of the assembled sequence, e.g., 4600000 for E.Coli
4.6Mb genome.

it is not explained in detail how this info is used and what happens if you are dealing with a new organism or mixed transcriptome sample/ environmental sample? Do we just throw in a value?

another question is how to I use > 4 cores using PBS to use multi threads?

[Rao]

based on genome size... i think it decides k-mer size that used for error correction

[javijevi]

refLength parameter

Maybe does it (also) use the refLength parameter for calculating the estimated frequency cutoff of trusted seeds in the spectrum construction step?

I think it is a very crucial data with a very strong influence in the quality of the results, and its use should be further clarified...

Does any developer can help?

[kevleb]

Does anybody have an idea about the interest to run saet tool concerning smallRNA sequencing ?
If yes what could be the "expected length of the assembled sequence" ?
For human sample it could be :
- 70.000 for mirs
- 8.000.000 for human ncRNAs
- 110.000.000 for refseq human mRNAs
- 2.900.000.000 for whole genome

[westerman]

An interesting question for which I have no answer. Since I have been wondering this myself let us hope that someone can answer. The entire SAET modification is a black-box mystery to me.

[KevinLam]

I think ABI finally bothered to release more information.
Basically it works like Softgenetics condensation tool.

it finds reads that have a single mismatch with all the similar reads and based on that it corrects the basecalls.

I leave it up to you to decide if this is a good thing

[westerman]

Since a single mismatch can only be a sequencing error then, in theory, correcting the error is a Good Thing. In practice I find that running SAET can take a long time. My last SAET took 100 CPU hours (about 12 hours wall time) but before that I ran a SAET that was heading off into multiple days of wall time. I eventually broke down the read file into two parts and ran SAET on each part. But that weakens the algorithm.

I would almost rather have the mapping part of the algorithm deal with single mismatch errors via throwing away the reads or doing as good of a mapping as it can. I might take my current project (a full plate, F3/R3 pairs so lots of data) and process it both with and without SAET to see if makes any difference.

My field specialist recommends running SAET only on small genomes (e.g., bacterial).

[schmima]

'expected length of the assembled sequence'

Maybe I'm wrong, but I understand it like this:
Assume you would use your reads (!) to assemble a genome/sequence (de novo). Then - how long would the sequence be? => use this as the parameter.
Ergo - in a nice world - if you sequence...
...genomic DNA use the genome size.
...mRNA use the transcriptome size (count only one gene variant per locus)
... and so on

Sounds somehow easy - but a 'near to correct' estimate seems - to me - hard to get in some cases... are the reads only from the type of nucleic acid you were interested in? I guess this is often not the case...
as an example - mRNA-seq with amplified material (poly-A or random priming):
=> origin of reads:
DNAse digest not 100% efficient (what is normally the case):
the expected length of the assembled sequence would increase (genomic contamination).
=> systematic coverage bias:
3' bias (for poly-A priming):
the expected length of the assembled sequence would drop (in some cases dramatically).
=> coverage in general:
assume you have reads of totally 1 gigabases from human genomic DNA. Using the human genome size would not be a good idea as you are not able to assemble 2.900.000.000 bases with 1 gigabases. And also 1 giga is not a good estimate as you would not expect a continuous coverage of 1.

Hm - for smallRNAs in your 'human'-case I guess that you would have to think about where your reads (! this is equal to small RNAs if you are sure that all reads are coming from small RNAs) could come from and use the size of this sequence.

However... I don't dare to give an estimate
and anyway: it's called estimate ^^

added later on:
From the bioscope manual (why don't they write this inside the SAET readme?):
'The expected length of the assembled sequence, for example: 4,600,000
for E.coli, 4.6 Mb genome, or 30,000,000 for Whole Human Transcriptome.'

So seems that I was not completely wrong...

I just still don't know what for the parameter is used during the correction...
checked SAET run with refLength = 120'000'000 compared to 13'000'000.
Number of changed reads:
refLength = 120'000'000: 19'703'256
refLength = 13'000'000: 23'926'281
total reads: 43'740'114
=> the parameter is definitely used in the calculation...
via taking into account the probability of random read-overlaps? or changed seedlength?

Anyone knows?

[poisson200]

Re-open the SAET debate

From the Bioscope 1.3 manual;
"The set of such trusted k-mers approximates the set of all k-mers in the
genome".....So

If I understand SAET, it is looking for K-mers that don't exist in a population of K-mers and corrects them (because they are not trusted)?

If you are working with reads from a genome or transcriptome, from a model species, that is going to be close to complete (like human or mouse); Why not use the transcriptome or genome to generate the model kmers?

It possibly could be this option, saet.outspecbin=1

Has anyone tried this? I am about to and not sure whether it is
a) possible
b) should I convert base space genome to colourspace

[golharam]

Is SAET still available on solidsoftwaretools? I don't seem to have access to it anymore.

[schmima]

Yes - but it's within the denovo2 package which you can download at:

http://solidsoftwaretools.com/gf/project/denovo/frs/

[jbdorr]

download denovo

denovo2 doesnt exist in the solid software tools page (http://solidsoftwaretools.com/gf/project/denovo/frs/).

can downloaded of my Master's Degree Page www.inf.ufpr.br/jbdorr/denovo2.tgz