Merging ssu-align alignments

Question

ssu-align is an HMM-based MSA tool for ribosome's small subunit (SSU) sequences. It's bundled with three SSU models:

• archaeal,
• bacterial
• eukaryotic

Approach Given a set of archaeal and bacterial sequences the tool will first identify the best model for each sequence and produce two separate alignments for archaea and bacteria, and the tool provides no merging functionality.

Rationale The underlying issue with SSU is the secondary structure, which results in: 1) indels; 2) pairwise point mutations (to maintain the secondary structure)

Problem While you can force it to use a single model for all sequences, aligning archaeal sequences to the bacterial model will produce poor results.

Solution? Nevertheless, the authors of GreenGenes claim that they've used ssu-align to produce two MSAs (one for archaea and bacteria) and build a joint phylogenetic tree using FastTree, but they say nothing about the way they've managed to merge the alignments. Am I missing some functionality from SSU-align?

Note Merging two giant MSAs using conventional tools, such as MAFFT, won't get me far.

P.S.

I've already tried contacting the authors of the GreenGenes paper, but the corresponding author hasn't responded.

Answer 1

I think this is more to do with tree building than alignment construction.

I believe the underlying issue is the 1) distribution of indels, this results in a blockwise deletion in the resulting phylogeny and 2) nucleotide saturation.

Indels Thus, for indels

Archaeal 
XXXX----YYYYY
bacterial
----ZZZZYYYYY

The resulting phylogeny will be based on YYYYY and XXXX is ignored.

The alternative explanation is that the genetic distances involved have resulted in saturation of the nucleotide sequences. SSU has to be a nucleotide phylogeny and the level of reversion mutations over a vast amount of evolutionary time can become indistinguishable from random.

With regards the first point, you can replace the indels "-" with "N". This is a cheat to trick a tree building program into not discarding the indel. Model based tree building programs have no ability to create a genetic distance between a base pair and an indel, so this data is essentially discarded from the tree. Worse it will remove the site from phylogenetic inference - which can be a massive loss of data. Two notes one this,

• Some max. likelihood, Bayesian tree programs can avoid blockwise deletion, I don't know programmatically how they do this;

• In PAML the distribution of indels was modelled, it doesn't appear to have been widely adopted however, but might be of value.

Alternative The alternative method is really for protein genes and called "RY" coding, it might work on SSU because of the compensatory mutations involved in maintaining a secondary structure of the SSU.

Word of warning, the saturation may be so great, very little nucleotide point mutation is preserved. Essentially we are talking about preservation of phylogenetic signal across the origins of life ... and that is a very, very long time ago.