RAxML-ng adaptive segmentation fault on Linux

[Kevin Myers]

I am running RAxML-NG v. 1.2.1-adaptive released on 11.04.2024 on an Intel Xeon Silver 4216 CPU @ 2.10GHz, 32 cores, 503 GB RAM. I have 2136 taxa and 5035 sites. I've tried to run RAxML-ng adaptive multiple times and every time, when RAxML-ng adaptive starts to run bootstrapping analysis, I get the following error (not in the log file, only printed to the screen):

/mnt/bigdata/linuxhome/kmyers/Jacob_RAxML/Raxml.sh: line 2: 84481 Segmentation fault      (core dumped)

I am running this on our server cluster using HTCondor, but it's assigned to the machine above. It is running Rocky 8.10 OS. I am not sure what to try. I am attaching the full debug log file. Any help would be greatly appreciated.

Thanks,

Kevin

[Anastasis Togkousidis]

Dear Kevin,

Thank you for reporting this bug. In fact, this version is quite old, and back in the day we simply merged the adaptive tree search into the main RAxML-NG code, without synchronizing it with other basic functionalities, such as bootstrapping.

There is the pre-released (beta) version of RAxML-NG v2.0, which combines everything. You can download the binary file for this version here:

https://github.com/amkozlov/raxml-ng/releases/tag/2.0-beta2

By default, this version executes the adaptive heuristic (see: https://academic.oup.com/mbe/article/40/10/msad227/7296053 ). In order to execute a similar command in this beta version, you need to run (based on the command that I extracted from your logfile):

./raxml-ng --all --msa gtdbtk.bac120.user_msa.fasta --model LG+G8+F --prefix Proteo_raxml --threads 62 --seed 2 --bs-trees 500 --log debug

Since your alignment comprises more than 2,000 taxa, you might consider using the --fast mode (it should be the default mode for large alignments, but in any case you can explicitly invoke it):

./raxml-ng --all --fast --msa gtdbtk.bac120.user_msa.fasta --model LG+G8+F --prefix Proteo_raxml --threads 62 --seed 2 --bs-trees 500 --log debug

Moreover, based on your sites-over-taxa ratio, I would expect the phylogenetic signal in your alignment to be somewhat weak. Interestingly, the predicted difficulty score is 0.31, and your analysis seems promising.

Finally, just in case you are thinking of using MPI, please keep in mind that early stopping (see: https://academic.oup.com/sysbio/advance-article/doi/10.1093/sysbio/syaf043/8154032 ) does not currently work with MPI. You can disable it by setting a fixed epsilon threshold (--lh-epsilon 10):

./raxml-ng --all --fast --msa gtdbtk.bac120.user_msa.fasta --model LG+G8+F --prefix Proteo_raxml --lh-epsilon 10 --threads 62 --seed 2 --bs-trees 500 --log debug

Please let us know how all this works for you.

Best, 
Anastasis

[Kevin Myers]

Thanks Anastasis!

We will try the beta version on our cluster and see how that works. I will reply later after we get it installed and we do a test run.

Appreciate the fast reply!

Kevin