[maker-devel] Doubts in the synthesis part of MAKER

[Carson Holt]

One more detail.

There are basically 5 steps per level.

Load -> this step creates the chunks for that level.  This is where I
decide how many chunks to make, and any special variables need to be
generated for packaging into that chunk.
Init --> these is just a declaration of the variables to package into a
chunk (only give the chunk what it needs)
Run --> these is the actual code that will be run after the chunk is
transported to it's destination
Result --> this describes how to merge results of that chunk back into the
parent object
Flow --> this decides what to do when all chunks for that level are
complete (i.e. which level to move onto next).  Default is next level in
linear succession, but it can jump forward and backwards several levels if
needed.

Thanks,
Carson


On 13-03-11 10:51 AM, "Carson Holt" <carsonhh at gmail.com> wrote:

>Hints are basically CDS location, exon location, and intron location.
>The 
>CDS hints are based on protein alignment.  Intron and exon hints are
>based 
>on the EST alignments, which when polished should give exact intron
>coordinates.  Ironically the most useless part of the gene model is
>actually the most informative feature for gene prediction (the intron
>coordinates).
>
>lib/Process/MPIchunk.pm will have the steps in the _go method.  It is a
>little hard to follow as MAKER is designed for distributed
>parallelization 
>(i.e. parallelization without shared memory with steps potentially
>divided 
>on different machines on the other end of the network).
>
>It is divided into MPItier and MPIchunk objects.  The MPItier object
>encapsulate a series of linear steps or 'levels' while the MPIchunk
>objects encapsulate a single step sent to a machine across the network
>and 
>it exists within a single 'level' of the MPITier object.  Note there can
>be multiple chunks assigned to a 'level'.  MPItiers can also have
>MPITiers 
>as children at a given level instead of MPIchunks, so the process
>structure then branches like a tree and can then merges back somewhere in
>the middle of the algorithm.
>
>The 'maker' script is really just the communication script for the
>objects.  In MPI one maker thread is launched to handle communication and
>another to run the MPItiers and MPIchunks.  They communication threads
>then pass MPIchunks and MPITiers back and forth across the network by
>either requesting things to do from other nodes or by asking for help if
>they have a large number of MPIChunks or MPItiers to process.
>
>Thanks,
>Carson
>
>
>
>
>
>On 13-03-11 10:33 AM, "Sitaram Rajaraman" <sitaram.rajaraman at helsinki.fi>
>wrote:
>
>>Hello MAKER developers,
>>			I'm Sitaram, working as a Bioinformatician at the University of
>>Helsinki. We are trying out MAKER as part of a gene
>>prediction/annotation
>>pipeline and have some doubts regarding this. In the synthesis step in
>>the paper, I find it a bit hard to visualise how the hints are generated
>>from the various sources and the scores are calculated. It would be nice
>>if you could throw some light on this. Also if you could point to the
>>particular .Pm file which contains the actual source code, it would be
>>convenient as there quite a lot of source code and debugging the whole
>>set is bit cumbersome.
>>
>>Regards,
>>
>>-- 
>>Sitaram Rajaraman,
>>Plant Stress Research Group,
>>Dept of Biosciences,
>>University of Helsinki.
>>
>>
>>_______________________________________________
>>maker-devel mailing list
>>maker-devel at box290.bluehost.com
>>http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org