[maker-devel] Incorporating cufflinks in maker

[Carson Holt]

BRAKER is basically a step by step running of Augustus.

1. Run GeneMark to get preliminary models to train Augustus (this is essentially the same as the CEGMA, est2genome, and protein2genome step use for training Augustus with MAKER).  You need something to start with even if Augustus is not perfectly trained.
2. Generate intron only hints from a BAM file (MAKER generates these from est2genome alignments of GFF3 alignments). This is basically a script in the Augustus distribution that turns BAM’s into intron hints (it’s basically cufflinks).
3. Rerun Augustus using the intron only hints.  The key here is that BRAKER generates a new hint weights file on the fly (file does not already exist in the Augustus config directory). It is tuned to give better performance than the default weights files.


Basically the difference between Augustus inside of BRAKER vs MAKER is that Mario has tuned the hints and weights file.  MAKER right now uses a default file in the Augustus config directory.  I’ve been thinking of just taking the BRAKER weights file and letting MAKER generate the same tuned weights on the fly, which would improve Augustus performance inside MAKER.  Mario also gave me some pointers on what hint types I should drop and add at PAG some time ago. Right now I use the intron_part hint inside of MAKER, and Mario says he gets better performance by using the exact match intron hint.  In fact that is the only hint type he uses in BRAKER.  He doesn’t use the CDS hint either.

Thanks,
Carson




> On Apr 3, 2015, at 2:27 AM, Holmer, Rens <rens.holmer at wur.nl> wrote:
> 
> Hi,
> 
> Great follow-ups! I indeed also use Genome-Guided Trinity to generate transcripts in parallel to cufflinks. I find that trinity denovo is actually often more noisy than cufflinks/tophat/transdecoder. For instance trinity often transcripts in intron regions because of spurious pre-mRNA reads.
> Currently I use BRAKER in stead of augustus, and the results are great. One thing to note is that augustus within maker does not provide the same results as the BRAKER output. I suspect this is because maker sends additional hints to the gene predictors on the fly. In my case BRAKER seems more accurate the the maker-augustus.
> 
> So I provide BRAKER output as pred-gff, transdecoder output as model-gff and trinity transcripts as EST.
> 
> Would love to hear comments!
> 
> Cheers,
> 
> Rens
> 
> On 03 Apr 2015, at 07:37, Jason Stajich <jason.stajich at gmail.com<mailto:jason.stajich at gmail.com>> wrote:
> 
> I also echo this - also has anyone given BRAKER a try as an alternative gene prediction input from augustus with hints through MAKER?
> http://bioinf.uni-greifswald.de/augustus/downloads/
> 
> 
> Jason Stajich
> jason.stajich at gmail.com<mailto:jason.stajich at gmail.com>
> 
> On Thu, Apr 2, 2015 at 10:36 AM, Carson Holt <carsonhh at gmail.com<mailto:carsonhh at gmail.com>> wrote:
> I actually find it better to process data with Trinity.  Cufflinks results tend to be too noisy, and any sensitivity lost by using an assembly approach instead of an alignment based approach is generally rescued by protein data.
> 
> —Carson
> 
> 
>> On Mar 30, 2015, at 12:12 AM, Holmer, Rens <rens.holmer at wur.nl<mailto:rens.holmer at wur.nl>> wrote:
>> 
>> Hi maker team,
>> 
>> I am currently working on a project where we want to incorporate quite a lot of RNA-seq into our annotation. Currently I see two options:
>> 
>> Provide the cufflinks output as EST-gff
>> Process cufflinks output with TransDecoder (find ORFs, annotate UTR, etc) and provide this as either pred_gff or model_gff
>> 
>> What would you suggest, and what would be the required formatting for both options?
>> 
>> Thanks in advance,
>> 
>> Rens Holmer
>> _______________________________________________
>> maker-devel mailing list
>> maker-devel at box290.bluehost.com<mailto:maker-devel at box290.bluehost.com>
>> http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
> 
> 
> _______________________________________________
> maker-devel mailing list
> maker-devel at box290.bluehost.com<mailto:maker-devel at box290.bluehost.com>
> http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
> 
>