[maker-devel] Help with updating an annotation

[Saad Arif]

Dear All,

I would like to use Maker pipeline to expand a current annotation (new isoforms and novel genes with respect to current annotation) and was wondering if anyone had experience with this and or suggestions to my questions.

Briefly:

I have tophat splice junctions from RNAseq data or alternatively cufflinks generated transcript models (fasts format) that i want to use as my new data (est_gff or est).

I want to provide the current Ensembl annotation for gene prediction but i want this annotation to remain unchanged. Hence, i’m not sure if i should provide this annotation as pred_gff
or model_gff. Can the model_gff be used for gene prediction or is this just a subset of pred_gff that remain unaltered? Can we provide the same annotation for both options (pred_ and mod_gff)?



Importantly, my main goal is to use the new RNAseq data to add more isoforms and (any) novel genes to the existing Ensembl annotation. Any thoughts or suggestions on how to go about this would be sincerely appreciated.


Thanks in advance,
saad




_______________________________________________
maker-devel mailing list
maker...@box290.bluehost.com
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org

[Carson Holt]

Use the cufflinks instead of the tophat features (tophat tends to be
really noisy). Give the existing models to model_gff (they will then
always be kept unless something better is found). There is no option to
keep models and then just add isoforms. The model_gff input will either
be kept as is (unchanged), or replaced with an updated model suggested by
the evidence (the updated model may contain multiple isoforms though), and
map_forward=1 can be used to pull names forward from the old model onto
the new models.

Thansk,
Carson

[Saad Arif]

Thank you for the response. I still have one question though, with these options:

est_GFF=cufflinksout.GFF

modle_GFF= ensembl reference.GFF

What happens to cufflinks assembled transcripts that are not confined to current gene loci (i.e. novel genes in cufflinks ouput)? Would i have to prepare ab initio gene predictions for each of these predicted 'new' genes?
Is there a simple way to combine adding (new genes) and improving of an existing annotation?

Any feedback on this would be greatly appreciated.

saad

[Daniel Ence]

Hi Saad, 

Maker doesn't view EST or protein evidence as a gene model in themselves. There's a good reason for this. Aligners like blast  don't guarantee complete gene models, with accurate start and stop codons and splice sites. With it's default settings maker won't make a gene model unless there's evidence that overlaps an ab-initio prediction (or something from the pred_gff option). 

You can use est2genome to promote everything from the est_gff option to a gene model, but this will probably give you many spurious results. What you're saying with est2genome is, "Everything that this tool found is a complete gene model." I don't think that's true even for cufflinks output. 

One of the gene predictors that can run internally is snap. It's really easy to train; here's a link to a tutorial for training it: http://weatherby.genetics.utah.edu/MAKER/wiki/index.php/MAKER_Tutorial_for_GMOD_Online_Training_2014#Training_ab_initio_Gene_Predictors

Let me know if that helps, or if you have more question

~Daniel

Daniel Ence

Graduate Student

de...@genetics.utah.edu
Eccles Institute of Human Genetics
University of Utah
15 North 2030 East, Room 2100
Salt Lake City, UT 84112-5330

On Jun 18, 2014, at 5:09 AM, Saad Arif <saad...@tuebingen.mpg.de>

[Daniel Ence]

Hi Saad, That seems to be right to me. You'll do one run of MAKER with the cufflinks output and est2genome turned on and train SNAP on that output. You can repeat this as many times as you want, but in my experience you don't gain much in predictive power beyond two rounds of training. 

Next, you'll turn on SNAP and turn off est2genome, but still include the cufflinks and proteome evidence and the ensemble models. The other ab initio predictors that maker can use internally (genemark and augustus) are worth looking into also. Genemark does a self-training thing, but can take a couple of days depending on how large your genome is. Augustus takes a lot of time and effort to train, but comes with many prebuilt training files. If one of its prebuilt files is close to your species of interest, you can just use that instead. 

~Daniel

Daniel Ence

Graduate Student

de...@genetics.utah.edu
Eccles Institute of Human Genetics
University of Utah
15 North 2030 East, Room 2100
Salt Lake City, UT 84112-5330

On Jun 18, 2014, at 10:42 AM, Saad Arif <saad...@tuebingen.mpg.de>

 wrote:

Thanks Daniel. I think it's more clear to me now.

So If I understand correctly now: I have to specify an ab initio gene model for any locus that I wish to annotate using evidence alignment (i.e. there must be a preexisting model)? These ab initio gene models can be trained internally in Maker with SNAP using my cufflinks output as EST evidence.Alternatively, I can provide alternative ab inito predictions (for regions not present in my ensembl ref passed to model_GFF) for regions overlapping my cufflinks output via the pred_GFF option? 

Since i'm interested in unannotated regions, i'm also passing in reference proteomes of closely related species as protein homology evidence.

As such i should be able to keep, only evidence supported predictions (for regions not present in my model_GFF and or better supported models for present regions) from my pred_GFF and merge them with Ensembl annotations from the model_GFF?

Let me know if i'm still missing something here. 

Thanks in advance.

best,

Saad

[Saad Arif]

Thanks Daniel. I think it's more clear to me now.

So If I understand correctly now: I have to specify an ab initio gene model for any locus that I wish to annotate using evidence alignment (i.e. there must be a preexisting model)? These ab initio gene models can be trained internally in Maker with SNAP using my cufflinks output as EST evidence.Alternatively, I can provide alternative ab inito predictions (for regions not present in my ensembl ref passed to model_GFF) for regions overlapping my cufflinks output via the pred_GFF option? 

Since i'm interested in unannotated regions, i'm also passing in reference proteomes of closely related species as protein homology evidence.

As such i should be able to keep, only evidence supported predictions (for regions not present in my model_GFF and or better supported models for present regions) from my pred_GFF and merge them with Ensembl annotations from the model_GFF?

Let me know if i'm still missing something here. 

Thanks in advance.

best,

Saad

On 18 Jun 2014, at 17:21, Daniel Ence wrote:

[Carson Holt]

"I have to specify an ab initio gene model for any locus that I wish to annotate using evidence alignment (i.e. there must be a preexisting model)?"

Not exactly.  You need to supply an HMM for SNAP or species file for Augusutus, etc.  MAKER doesn't generate gene predictions, SNAP does. You cannot get updated models unless you've provided a way for those models to be updated.  MAKER will provide SNAP/Augustus with hints to make them perform better based on the evidence, but those hints won't even be genertated and the programs won't even run unless you provide the HMM.  Also if you provide models in gff3 format to pred_gff, there is not hint feedback (because there is no program to receive the hints - just an immutable GFF3 file).

If you don't have an HMM for SNAP for your organism, you can generate one using the documentation here (from GMOD 2014 tutorial) --> http://weatherby.genetics.utah.edu/MAKER/wiki/index.php/MAKER_Tutorial_for_GMOD_Online_Training_2014#Training_ab_initio_Gene_Predictors

--Carson

[Daniel Ence]

Hi Saad, I think that's correct. As a sub step for each of the steps you listed, I would also choose one or two large scaffolds out of your assembly to use as a test set and use that test set to make sure that all you are getting output like you'd expect, before running MAKER on the whole genome. 

Let me know if there's anything else we can do to help. 

~Daniel

Daniel Ence

Graduate Student

de...@genetics.utah.edu
Eccles Institute of Human Genetics
University of Utah
15 North 2030 East, Room 2100
Salt Lake City, UT 84112-5330

On Jul 7, 2014, at 7:08 AM, Saad Arif <saad...@tuebingen.mpg.de>

 wrote:

Thanks for this. Would the following protocol be appropriate then, given i want to augment and merge an existing annotation with any novel genes:

i) Run MAKER pipeline iteratively to generate an HMM for SNAP using my new RNAseq data and protein fastas from closely related organisms (with esttogenome and proteintogenome options on).

ii) Turn off esttoGenome and proteintoGenome options and run Maker with my RNAseq evidence, protein fastas, SNAP HMM and my current annotation as model_GFF.

Thanks in advance for any input.

best,

Saad

[Saad Arif]

Thanks for this. Would the following protocol be appropriate then, given i want to augment and merge an existing annotation with any novel genes:

i) Run MAKER pipeline iteratively to generate an HMM for SNAP using my new RNAseq data and protein fastas from closely related organisms (with esttogenome and proteintogenome options on).

ii) Turn off esttoGenome and proteintoGenome options and run Maker with my RNAseq evidence, protein fastas, SNAP HMM and my current annotation as model_GFF.

Thanks in advance for any input.

best,

Saad

On 20 Jun 2014, at 23:42, Carson Holt <cars...@gmail.com> wrote: