[maker-devel] About split genes in MAKER annotation
Carson Holt
carsonhh at gmail.com
Wed Dec 19 10:14:25 MST 2018
Given how you have plenty of protein evidence aligning well, I would suggest you do protein2genome only and not est2genome to build your training set (your est2genome results are more fragmented). You can further filter for canonical start and stop codons as well as protein completeness which will be in the score column in the GFF3.
Once trained, you run maker again with augustus_species= set to your newly trained species. MAKER will run augustus for you and generate the hints for augustus using the est= and protein= files you provide.
—Carson
> On Dec 18, 2018, at 8:20 PM, Prashant Narendra SHINGATE <prashantns at imcb.a-star.edu.sg> wrote:
>
> Dear Carson,
>
> Thanks for clarifying that I should not base gene models on evidence based prediction but train AUGUSTUS. I will carry out AUGUSTUS training using rough gene models predicted in evidence based run and also follow entire annotation protocol. I am assuming that even though Exonerate splits genes based on aligned ESTs/proteins, AUGUSTUS will be able to predict full-length genes.
>
> I need your suggestion on AUGUSTUS training. We have ~10,000 full-length transcripts and corresponding proteins (besides a large number of fragments) generated by assembling RNAseq transcripts from the same individual used for genome sequencing. However, I understand from AUGUSTUS training tutorial that ~1000 gene models are good enough to train AUGUSTUS. So is it good idea to choose rough gene models predicted based on 10,000 full-length transcripts/proteins for training AUGUSTUS or should I randomly pick 1000 gene models from the entire evidence based run?
>
> Also about hint files used as input for AUGUSTUS, should I include BLAST alignments of transcripts (BLASTN) and proteins (BLASTX) in hint files or only exonerate alignments are recommended in hint files? Please clarify.
>
> Thanks once again for your help and time.
>
> Best Regards,
>
> Prashant Shingate, PhD <mailto:prashantns at imcb.a-star.edu.sg> :: Research Fellow :: Comparative and Medical Genomics Lab :: Institute of Molecular and Cell Biology (IMCB) :: Agency for Science, Technology and Research (A*STAR)
> 61 Biopolis Drive :: #05-04 Proteos :: Singapore 138673 :: DID (+65) 6586 9570 <tel:(+65)%206586%209570> :: Fax (+65) 6779 1117 <tel:(+65)%206779%201117>:: http://www.imcb.a-star.edu.sg/ <http://www.imcb.a-star.edu.sg/>
> We advance science and develop innovative technology to further economic growth and improve lives.
>
>
> From: Carson Holt [mailto:carsonhh at gmail.com <mailto:carsonhh at gmail.com>]
> Sent: Tuesday, 18 December, 2018 11:15 PM
> To: Prashant Narendra SHINGATE
> Cc: Byrappa VENKATESH; maker-devel at yandell-lab.org <mailto:maker-devel at yandell-lab.org>
> Subject: Re: About split genes in MAKER annotation
>
> You are using est2genome and protein2genome. It is not doing gene prediction, rather it’s just tiling EST’s or trying to turn protein alignments directly into rough models to be used as training sets. That is why the gene is split, because there is no long transcript alignment, just two alignments that are cut and pasted directly from exonerate down onto the assembly. You should not use est2genome or protein2genome as final models. You need to train SNAP or Augustus.
>
> See Documentation —>
> http://weatherby.genetics.utah.edu/MAKER/wiki/index.php/MAKER_Tutorial_for_WGS_Assembly_and_Annotation_Winter_School_2018#Training_ab_initio_Gene_Predictors <http://weatherby.genetics.utah.edu/MAKER/wiki/index.php/MAKER_Tutorial_for_WGS_Assembly_and_Annotation_Winter_School_2018#Training_ab_initio_Gene_Predictors>
>
> —Carson
>
>
>
> On Dec 18, 2018, at 12:23 AM, Prashant Narendra SHINGATE <prashantns at imcb.a-star.edu.sg <mailto:prashantns at imcb.a-star.edu.sg>> wrote:
>
> Hi Carson,
>
> Thank you for the reply.
>
> Please find attached the GFF file of Scaffold61 along with related ctl files. The coordinates of the gene I am referring to is from 698,453 to 840,581 bp. Several proteins, short and long are aligned to this locus including a 1200aa protein and a 2,069 aa (XP_022239675.1) protein. The latter is aligned completely to scaffold61 from 698,453 to 840,581 bp with >90% identity. Please see the related line to this alignment from GFF file.
>
> SCAffold61 blastx protein_match 698453 840581 730 - . ID=SCAffold61:hit:1861:3.10.0.0;Name=XP_022239675.1
>
> However, in evidence-based run, this gene is split into two fragments. Please see the related lines from GFF file as follow:
>
> SCAffold61 maker gene 708052 717805 . - . ID=maker-SCAffold61-exonerate_est2genome-gene-0.95;Name=maker-SCAffold61-exonerate_est2genome-gene-0.95
> SCAffold61 maker gene 748651 770415 . - . ID=maker-SCAffold61-exonerate_est2genome-gene-0.96;Name=maker-SCAffold61-exonerate_est2genome-gene-0.96
>
> It looks like Exonerate prediction is based only on ESTs which are fragmented and the full-length protein aligned to this locus is completely ignored. We have seen this type of priority for ESTs in other loci also resulting in split gene prediction (sometime 3 to 4 fragments) in spite of alignment of longer full-length proteins to the assembly. Our ESTs (Trinity assembled RNAseq transcripts) were generated from the same individual whose genome was sequenced (and hence the identify is close to 100%). If we align only proteins, Exonerate still splits the gene based on shorter proteins aligned to the locus. I would really appreciate if you can help us to solve this splitting of genes despite alignment of full-length proteins to the assembly.
>
> I will be glad to send all reference protein and transcript sequences used for annotation, if required.
>
> Thanks for your time and help.
>
> Best regards,
>
> Prashant Shingate, PhD <mailto:prashantns at imcb.a-star.edu.sg> :: Research Fellow :: Comparative and Medical Genomics Lab :: Institute of Molecular and Cell Biology (IMCB) :: Agency for Science, Technology and Research (A*STAR)
> 61 Biopolis Drive :: #05-04 Proteos :: Singapore 138673 :: DID (+65) 6586 9570 <tel:(+65)%206586%209570> :: Fax (+65) 6779 1117 <tel:(+65)%206779%201117>:: http://www.imcb.a-star.edu.sg/ <http://www.imcb.a-star.edu.sg/>
> We advance science and develop innovative technology to further economic growth and improve lives.
>
>
> From: Carson Holt [mailto:carsonhh at gmail.com <mailto:carsonhh at gmail.com>]
> Sent: Tuesday, 18 December, 2018 1:38 AM
> To: Prashant Narendra SHINGATE
> Cc: Byrappa VENKATESH; maker-devel at yandell-lab.org <mailto:maker-devel at yandell-lab.org>
> Subject: Re: About split genes in MAKER annotation
>
> It’s best to look at these in a browser like Apollo where you can also manipulate the intron/exon structure. What you will often find is that there is something that breaks the ORF or breaks splicing, so the predictors can’t build an end to end model even with the hints given. If you have a GFF3 just for the contig, I can also look at it in a browser to help point out the logic that lead to the model.
>
> —Carson
>
> On Dec 12, 2018, at 3:29 AM, Prashant Narendra SHINGATE <prashantns at imcb.a-star.edu.sg <mailto:prashantns at imcb.a-star.edu.sg>> wrote:
>
> Hi Carson,
>
> I am Prashant a Bioinformatics postdoctoral fellow from Prof B Venkatesh’s lab, IMCB, A*STAR. I am using MAKER-tool to annotate an invertebrate genome (~2Gb). During annotation process, we found several instances of split geneseven though we have full-length reference protein sequences from very closely related species. Hence we decided to look at one of the loci to understand the reason behind it and to optimize the parameters.
>
> We looked at a gene ~110kb long and codes for a ~1200 amino acid protein. We have a highly identical reference protein (>90% identity and 100% coverage) from another species. In addition we also have a high coverage Trinitytranscript assembly from our species. Still, this gene is split into 4 fragments during evidence-based MAKER run. On closer a look, we found that the above mentioned closely related protein is not aligned by exonerate (protein2genome) even though it is the closest protein to this gene in our dataset. It looks like the program is giving more weightage to transcripts which are typically fragments of the gene. So we are at a loss as to how to predict this gene in full.
>
> For your reference, I am herewith enclosing maker_opts.ctl file and maker_bopts.ctl. I will be glad to share the scaffold sequence and other input files if required.
>
> Can you please help me to understand the reason behind MAKER not able to use the full-length reference protein for gene prediction and how we can overcome this problem.
>
> Thanks for your time and help.
>
> Best regards,
>
> Prashant Shingate, PhD <mailto:prashantns at imcb.a-star.edu.sg> :: Research Fellow :: Comparative and Medical Genomics Lab :: Institute of Molecular and Cell Biology (IMCB) :: Agency for Science, Technology and Research (A*STAR)
> 61 Biopolis Drive :: #05-04 Proteos :: Singapore 138673 :: DID (+65) 6586 9570 <tel:(+65)%206586%209570> :: Fax (+65) 6779 1117 <tel:(+65)%206779%201117>:: http://www.imcb.a-star.edu.sg/ <http://www.imcb.a-star.edu.sg/>
> We advance science and develop innovative technology to further economic growth and improve lives.
>
>
>
>
> Note: This message may contain confidential information. If this Email/Fax has been sent to you by mistake, please notify the sender and delete it immediately. Thank you.
> <maker_opts.ctl><maker_opts.ctl>
>
>
>
> Note: This message may contain confidential information. If this Email/Fax has been sent to you by mistake, please notify the sender and delete it immediately. Thank you.
> <maker_bopts.ctl><maker_opts.ctl><SCAffold61.zip>
>
>
>
> Note: This message may contain confidential information. If this Email/Fax has been sent to you by mistake, please notify the sender and delete it immediately. Thank you.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://yandell-lab.org/pipermail/maker-devel_yandell-lab.org/attachments/20181219/cd2ed11f/attachment-0003.html>
More information about the maker-devel
mailing list