Maker crash with big protein evidence ?

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

Maker crash with big protein evidence ?

Patrick Tran Van-2

Hi,


I run maker with mpi and default option.

It works well with swissprot/uniprot as protein evidence but I wanted to provide a larger database (nr) and it run for a while but for longer scaffold it crashs after this :


in cluster::shadow_cluster...
 sorting hits in shadow cluster...
 j_size:3   current j:0
 j_size:3   current j:1
 j_size:3   current j:2
...finished clustering.


Did you already face something similar ?


Thanks.


Patrick 

_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
Reply | Threaded
Open this post in threaded view
|

Re: Maker crash with big protein evidence ?

Carson Holt-2
You may be running out of RAM. You can try setting depth_blast= parameters in maker_bopts.ctl to something like 10 or 20 to throw away more redundant data. But I would also suggest not using NR. It’s not curated to the extent swiss-prot is, so you will get a lot of poor models. The result will be a lot of bad false hints sent to the predictors (trash in trash out).

—Carson


On Jan 15, 2019, at 2:54 AM, Patrick Tran Van <[hidden email]> wrote:

Hi,

I run maker with mpi and default option. 
It works well with swissprot/uniprot as protein evidence but I wanted to provide a larger database (nr) and it run for a while but for longer scaffold it crashs after this :

in cluster::shadow_cluster...
 sorting hits in shadow cluster...
 j_size:3   current j:0
 j_size:3   current j:1
 j_size:3   current j:2
...finished clustering.

Did you already face something similar ?

Thanks.

Patrick 


_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
Reply | Threaded
Open this post in threaded view
|

Re: Maker crash with big protein evidence ?

Patrick Tran Van-2

Thanks Carson.

I thought also to use only swissprot at the beginning but I am working with a non model organism and I am scared that swissprot is not enough complete to capture all the gene prediction.


but if I use nr and I select only gene with AED < 0.5, it should be good no ? or do you have an AED cut off to recommend ?


Cheers.


Patrick

From: Carson Holt <[hidden email]>
Sent: Thursday, January 17, 2019 10:31:17 PM
To: Patrick Tran Van
Cc: [hidden email]
Subject: Re: Maker crash with big protein evidence ?
 
You may be running out of RAM. You can try setting depth_blast= parameters in maker_bopts.ctl to something like 10 or 20 to throw away more redundant data. But I would also suggest not using NR. It’s not curated to the extent swiss-prot is, so you will get a lot of poor models. The result will be a lot of bad false hints sent to the predictors (trash in trash out).

—Carson


On Jan 15, 2019, at 2:54 AM, Patrick Tran Van <[hidden email]> wrote:

Hi,

I run maker with mpi and default option. 
It works well with swissprot/uniprot as protein evidence but I wanted to provide a larger database (nr) and it run for a while but for longer scaffold it crashs after this :

in cluster::shadow_cluster...
 sorting hits in shadow cluster...
 j_size:3   current j:0
 j_size:3   current j:1
 j_size:3   current j:2
...finished clustering.

Did you already face something similar ?

Thanks.

Patrick 


_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
Reply | Threaded
Open this post in threaded view
|

Re: Maker crash with big protein evidence ?

Carson Holt-2
NR will likly create problems on more genes than it will help recover in comparison to swiss-prot. The AED cutoff can be a convenient way to remove models when you expect too much or poor quality evidence (but it’s more of a sledgehammer than a scalpel).  So using NR to get better sensitivity and then using AED to reduce sensitivity will likely just make all models worse in general while not adding many missed models.

—Carson



On Jan 18, 2019, at 10:09 AM, Patrick Tran Van <[hidden email]> wrote:

Thanks Carson. 
I thought also to use only swissprot at the beginning but I am working with a non model organism and I am scared that swissprot is not enough complete to capture all the gene prediction.

but if I use nr and I select only gene with AED < 0.5, it should be good no ? or do you have an AED cut off to recommend ?

Cheers.

Patrick

From: Carson Holt <[hidden email]>
Sent: Thursday, January 17, 2019 10:31:17 PM
To: Patrick Tran Van
Cc: [hidden email]
Subject: Re: Maker crash with big protein evidence ?
 
You may be running out of RAM. You can try setting depth_blast= parameters in maker_bopts.ctl to something like 10 or 20 to throw away more redundant data. But I would also suggest not using NR. It’s not curated to the extent swiss-prot is, so you will get a lot of poor models. The result will be a lot of bad false hints sent to the predictors (trash in trash out).

—Carson


On Jan 15, 2019, at 2:54 AM, Patrick Tran Van <[hidden email]> wrote:

Hi,

I run maker with mpi and default option. 
It works well with swissprot/uniprot as protein evidence but I wanted to provide a larger database (nr) and it run for a while but for longer scaffold it crashs after this :

in cluster::shadow_cluster...
 sorting hits in shadow cluster...
 j_size:3   current j:0
 j_size:3   current j:1
 j_size:3   current j:2
...finished clustering.

Did you already face something similar ?

Thanks.

Patrick 


_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org