Runtime of muscle -refine

Question

[jdru]

I have run muscle with the -refine flag on an alignment of ~25k mitochondrial sequences (each around 17k in length). Unfortunately even on verbose mode I am seeing no progress being outputted so I have no sense of how long it will take and whether it will be feasible. Does anyone know approximately what the runtime is for this option?

Thanks!

Answer 1

[Michael]

25k is a lot of sequences, what do you want to do after that, phylogeny? I think muscle is not ideal for so many sequences, you should try MAFFT instead, with some speed oriented parameter settings it is said to be able to align thousands of sequences. The manual recommends for up to 50k sequences:

   mafft --retree 1 --maxiterate 0 --nofft --parttree input [> output]

Also, adding --thread -1 migth use your CPUs efficiently.

https://mafft.cbrc.jp/alignment/software/manual/manual.html

It might still use a lot of memory, so look for other tools. There has a couple of questions about MSA of thousands of sequences, and the outcome seems to be that it is difficult to pull of without a high-mem server.

[jdru]

Thank you! Actually the alignment I have already was done with MAFFT - my problem is that I'm not sure how well MAFFT can get the indels right and that is critical for my project. So I was hoping muscle might be able to improve the existing alignment. But I suppose maybe it is not possible for an alignment of this size.

[Michael]

MAFFT has more options than MUSCLE to trade quality for speed and vice-versa. So, you can try to gradually move towards quality parameters, in particular increase maxiterate and retree. Also, just to be sure, I would check for 100% identical sequences. I haven't tried to align that many sequences but in general the alignments made by mafft were at least as good or better than the muscle alignments, when evaluating the resulting phylogenetic trees.