Internal error in ComputeBranchCache

[legason]

Dear Hyphy developers and users! ,

I tried to analyse my DNA sequences ~ 135kb long and 54 in total for recombination using the command below ENV=TOLERATE_NUMERICAL_ERRORS=1 hyphy gard --alignment EBV_trimmed.fasta --tree EBV_trimmed.fasta.treefile --max-breakpoints 500 --model GTR+G --cpu 8

Of course, this is after trying to construct the tree in guard, and it failed several times. So I constructed the tree in raxml and hoped running guard with a precomputed tree would make the analysis more efficient because HyPhy wouldn’t have to infer the tree dynamically. Is my alignment too large? I even tried to reduce the breakpoints to 100 but still got the same error. I appreciate your help.

Thanks, Ismail.

[spond]

Dear @legason,

A couple of things.

1. The ENV statement is a command argument to hyphy (not a shell var); see below
2. I would suggest using HYPHYMPI. Also the cpu= flag doesn't do anything (it's CPU=X, case sensitive)
3. The --tree option is ignored. It makes no sense for gard and --model is similarly not an option HyPhy understands

So something like

mpirun -np N HYPHYMPI gard --alignment EBV_trimmed.fasta  ENV="TOLERATE_NUMERICAL_ERRORS=1;" 

For long alignments like yours, 135kbp, I would expect GARD to run a loing time; it's not really designed for very long genomes (more like RNA viruses, or fragments), something on the order of seversl kbp.

Best,
Sergei

[legason]

Thanks, Sergei!

However, does supplying precomputed tree work or not?

[stevenweaver]

Dear @legason,

No, supplying a precomputed tree does not work for GARD. As mentioned in the response by @spond, the --tree option is ignored in GARD because the method itself is designed to infer trees dynamically as part of its recombination detection process. Using a precomputed tree will not make the analysis more efficient, as GARD still needs to evaluate different topologies at potential breakpoints to detect recombination events.

Best,
Steven

[legason]

Thanks, @spond @stevenweaver, for your responses. I have run it as per @spond guidance, but it takes hours.

Thanks again,
Ismail

[spond]

Dear @legason,

If you have highly similar sequences, doing a data compression like desciribed in #1448

Best,
Sergei

[legason]

Dear @spond
I will try it if the current session takes another 12 hours.

Thanks,
Ismail