Hello,

I'm trying to implement a PAML M8a model via Hyphy using the NielsenYang.bf batch file. However, I keep getting the message "Abort Trap: 6" when running. The errors.log and messages.log remain empty, as well.

I have tried to use both the direct command line interface as well as an inputRedirect, and both yield the error. I'm attaching my file for inputRedirect here. I've tried it with a variety of alignments/trees, all of which are fasta and newick, respectively, and the abort trap keeps occurring.


Any thoughts would be much appreciated.

Thanks,
Stephanie Spielman

---

An update - that batch file seems to work fine on Linux, but not on Mac OSX.

Hi Stephanie,

I haven't seen this behavior on OS X before: could you provide more details for me?

1). Which version of HyPhy are you running (command line, GUI, version number)?
2). What is the OS X version?
3). How much of the analysis finishes before you see the Abort Trap raised?

Sergei

Hi Sergei,

The time before the Abort Trap occurs varies - sometimes it occurs after only 3 seconds, sometimes 1-2 minutes, but usually by that time frame. The errors.log and messages.log are always empty, so I have no real idea of how much analysis is being run.

I'm using command line - HYPHY 2.1120130307beta(MP) for Darwin on x86_64 . OSX version is 10.8.4.

At this point, I'm running all analyses on Linux to avoid the errors, and it runs perfectly well on Linux.

However, I misspoke before - I am trying to replicate PAML's M8 model, "Beta & w." As far as I can tell, this model should only have a single category of omega>1 as the rest of the rates are from a beta distribution. However, many of my result files are coming out with two or three categories of omega>1 (rest is purifying). Is there a way I can prevent this from happening so that only one category is above 1?

Thanks,
Stephanie

Hi Stephanie,

Your HyPhy version seems a bit out of date -- please use the current 2.2 version (Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login). There have been occasional crashes on OS 10.7 and 10.8 due to some multithreading issues. Try including the CPU=1 argument on the command line: this will force HyPhy to run in single threaded mode, and if the issue no longer occurs, then the problem is due to multithreading (this issue is not limited to HyPhy, btw, but many packages that use OpenMP).

Beta & omega should indeed have one category greater than 1 (which can actually lead to bad statistical performance, see Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login. If you see more than one omega > 1 (presumably the others are close to 1?), then this is a numerical issue with HyPhy, but also indicative of the fact that beta & omega is a bad fit to your data (i.e. the beta distribution gets 'bunched' on the boundary). Are the p and q parameters high or low?

Unless you have compelling reasons to use Beta & omega, I would suggest you use FUBAR instead: it's much faster and more reliable: Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login

Sergei

Hi Sergei,

I am also using FUBAR in addition to this model, in a partial (although not primary) effort to compare the two. (FYI  - I am one of Claus's students and had the opportunity to speak with Konrad several months back when he visited UT, so FUBAR is definitely on the radar over here!)

While I have no doubt that the Beta+w omega is not the best fit to the data, I still am concerned about the performance of your implementation of the PAML model. All but one of the omega ratios it computes should be constrained to be <1, as that they are drawn from a beta distribution. Given this model constraint, it should be impossible to recover multiple categories of positively selected sites. I think it might be issue worth looking into, because even though I agree that the model is probably a bad fit, it should still be constrained to result in only one category of positively selected sites, with the rest purifying. In my mind, poor model fit should not result in "weird" values, but not values which violate the actual model constraints.

Thanks,
Stephanie

Hi Stephanie,

I agree with your sentiment on PAML's models implementations; I think what you see is a numerical artifact due to a 'weird' beta distribution shape. Can you send me the model fit that HyPhy writes out as a part of the analysis, so that i trouble-shoot the problem?

Sergei