[maker-devel] different exons predicted in different maker rounds

[Carson Holt]

Some of your cufflinks evidence is contradicting the existence of the exon. When you have alt_splice=0, the evidence is passed in in it’s entirety, and all sources are equal.  When you have alt_splice=1 set, certain pieces of spliced evidence are given higher priority (in an iterative fashion), and cannot be overridden by contradictory evidence.  The result is that there is a specific combination of evidence based hints that allow the gene predictor to find the exon, but when it sees everything the HMM doesn’t score the exon as very likely.  I’d recommend not running with cufflinks because its result usually have low specificity, so it generates a lot of bad hints.  Use Trinity instead to assembly everything, then allow it to be aligned inside of MAKER. Trinity assembled contigs give much greater specificity in the final results.

—Carson


> On Oct 5, 2015, at 6:06 AM, roscito <roscito at mpi-cbg.de> wrote:
> 
> Dear all,
> 
> First of all, I'd like to thank everyone in this forum for all the tips and comments on the best strategies for running MAKER, they have been really helpful so far.
> However, I still don't fully understand the behaviour of MAKER when ran iteratively, and I compare the predictions from each round. Let me explain:
> 
> My input data are the following:
> - the repeat-masked genome of a vertebrate (~2Gb);
> - mRNA data for this species mapped to the genome with tophat2 and assembled into transcripts with cufflinks;
> - exonerate-mapped proteins in gff3 format to the reference genome, from closely related species (global alignment)
> 
> For the first round of MAKER, I provided both cufflinks and exonerate-mapped proteins with the options est2genome and protein2genome = 1. From maker output, I generated the SNAP .hmm file (as the instructions in http://gmod.org/wiki/MAKER_Tutorial <http://gmod.org/wiki/MAKER_Tutorial>) and provided it as input to the second round of MAKER.
> For this second round I still gave cufflinks + exonerated proteins, but switched both est2genome ad protein2genome to 0. After finished, I generated SNAP .hmm once more and provided it for the 3rd and final round of MAKER, along with cufflinks and exonerated-mapped prots and est/prot2genome=0
> 
> As sort of a sanity check, I went on and ran a 4th round of MAKER with the SNAP .hmm file from round3, cufflinks and exonerated-mapped prots and est/prot2genome=0, and this time specifying alt_splice=1.
> For all the rounds, I also specified single_exon=1.
> 
> 
> I loaded the gene predictions from each round plus the cufflink transcripts and the exonerated proteins to the genome browser to visually inspect the output. I saw a few strange cases where MAKER doesn't seem to use the protein/mRNA evidences for the gene predictions, and I would greatly appreciate any feedback/ideas on what I could possible be doing wrong. Here are a few screenshots so you know what I'm talking about:
> 
> In this first example, MAKER misses a conserved exon for which there is both protein and mRNA evidence, and only if I specify alt_splice I get the exon 'back'.
> 
> <example1.png>
> 
> In this second example, MAKER completely ignores lots of exons, all conserved across vertebrates, and supported by protein/mRNA evidence.
> 
> <example2.png>
> 
> In the third example, there is no prediction from round1, the one from round2 matches the protein/mRNA evidence, and then in the final round3 and 4, an extra exon appears.
> 
> <example3.png>
> 
> 
> (hope you'l be able to see the images above)
> As I said, I would greatly appreciate any feedback on these strange cases. Perhaps I'm missing some parameter(s)?
> 
> Thanks a lot.
> All the best,
> Juliana
> _______________________________________________
> maker-devel mailing list
> maker-devel at box290.bluehost.com
> http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://yandell-lab.org/pipermail/maker-devel_yandell-lab.org/attachments/20151009/3e7a0444/attachment-0003.html>