Re: [Cg-pipeline-users] prediction - blastSeqs

[Lee K. <ls...@gm...>]

Sure.  When developing the strategy for gene prediction, we decided to
accept the longest orf possible.  Therefore, we always capture the whole
gene (if the locus is correctly identified as a gene because there are
undoubtedly some false positives), but the 5' region might be a false
positive.  In other words, a downstream start codon might be the true start
in CGP-predicted genes.  I believe in our publication, we correctly
identified genes 95% of the time, but we predicted the correct start 85% of
the time (and the true start site would have been downstream in 15% of the
predictions).


On Tue, Jul 2, 2013 at 5:14 AM, kokwei <kok...@gm...> wrote:

>  Thanks for your information. I am thinking the reference proteins I used
> as database are conserved in their gene model, so hoping that with the
> blast result against database will predict the "real start" (with
> reference) instead of based on ab initio which might have false prediction
> for start codon. With the current protein blast alignment, it will pinned
> to the closest start codon upstream, but the "real start" might be even
> more upstream.
>
> Can I have your opinion on this based on your experience with the
> optimization? Thanks.
>
>
> On 2/7/2013 3:36 AM, Lee Katz wrote:
>
> It looks like you are correct.  Over the last couple of years, I optimized
> a few different things and I guess I obviated that subroutine!
>
>  Yes, run_prediction will run GeneMark and Glimmer3, and combine with
> blast, making a high-confidence set of genes.  To see a graphical
> representation and explanation of our combining strategy, please see our
> 2010 paper:
> http://bioinformatics.oxfordjournals.org/content/26/15/1819.full
>
>
> On Mon, Jul 1, 2013 at 12:49 PM, kokwei <kok...@gm...> wrote:
>
>>
>> Hi,
>>
>> I noticed that during execution of pipeline with command <run_pipeline
>> predict -p project>, the module "sub blastSeqs($$$) {" at line 1817 is
>> not executed. Is that normal or only happen in my case? Attached the log
>> of the prediction for your reference.
>>
>> I guess that module will do the blast of the contig/scaffolds sequences
>> against the database selected and output the homology based gene model
>> to combine with other ab initio gene model by glimmer and genemark,
>> which is what I need and hope that this can be solved.
>>
>> Thanks.
>>
>>
>>
>>
>>
>>
>
>
>  --
> Lee Katz, Ph.D.
>
>
>


-- 
Lee Katz, Ph.D.