Comments on: Uncertainty in inferred alignments

By: Mailund on the Internet » Blog Archive » StatAlign: a new statistical alignment tool

Wed, 08 Oct 2008 17:25:15 +0000

[...] For a good introduction to the problems (and how statistical approaches to alignment can help), you should read Lunter et al. Uncertainty in homology inferences: Assessing and improving genomic sequence alignment Genome Res. 18:298-309, 2008 (or my summary of it here). [...]

By: Thomas Mailund

Thomas Mailund — Tue, 04 Mar 2008 11:11:28 +0000

That story gets better with beer as well ;-)

By: Roald Forsberg

Roald Forsberg — Tue, 04 Mar 2008 10:41:13 +0000

Thomas, I knew of the thoughts behind it but enjoyed reading your interpretation none the less. Also, when Dr Lunter explained this to me, beer might have been involved ;-) Look forward to hearing the story from Ox!

By: Thomas Mailund

Thomas Mailund — Tue, 04 Mar 2008 08:56:33 +0000

Roald, I thought you already knew about this paper ;-) Didn’t you discuss it with Gerton when he visited this summer?

By: Roald Forsberg

Roald Forsberg — Tue, 04 Mar 2008 06:15:03 +0000

Great post, I really enjoyed it!

By: Thomas Mailund

Thomas Mailund — Tue, 04 Mar 2008 04:04:42 +0000

Right, I’ll ask him, then :)

By: Reed A. Cartwright

Reed A. Cartwright — Mon, 03 Mar 2008 19:54:47 +0000

There is a serious interpretation issue with geometric gap models: they are not evolutionarily conserved. Human-Chimp, Rat-Mouse, and Human-Mouse comparisons can have the same underlying indel distribution model, but the geometric model can claim that they have different distributions. This is due to certain properties of the natural gap length distribution. I’m not going to give away the entire story here, but Asger may remember it.

By: Thomas Mailund

Thomas Mailund — Mon, 03 Mar 2008 17:48:32 +0000

That sounds interesting Reed. When you write up your results, I’d love to have a look.

I have only recently started getting interested in this particular problem — there were so many people working on it while I was in Oxford (and they seemed to have a lot of problems with it) that I didn’t really want to get involved. I currently have a paper in review where we combine a selection estimation model with an HMM and the EM algorithm (Baum-Welsh). Here we see an improvement in estimation accuracy compared to first estimating an alignment and then estimating parameters based on that.

I’m not sure I understand the problem with a geometric gap model… is the problem that gap lengths are not geometrically distributed? Is there really any reason to think they should be, except that it is easier to model geometrical distributions in HMMs? ;-)

I suppose there are good reasons why it could be geometrically distributed, but it really depends on the molecular mechanism that causes the indels.

Do you see any difference between areas close to genes — expected to be under selection — and sites that should have “neutral” indel rates?

What about the distribution between indels? Gerton Lunter had a paper a few years back — I think it was in PLoS Biology — about “the neutral indel rate” that was based on the distribution of distances between indels. Co-authors were Chris Ponting and Jotun Hein.

By: Reed A. Cartwright

Reed A. Cartwright — Mon, 03 Mar 2008 15:59:08 +0000

I work on alignment uncertainty, especially with respect to picking the best alignment models. I use a hidden markov model and an EM algorithm to sum over all possible alignment and estimate model parameters. I should write it up this semester, but my PI and I are working on a different project right now.

In my research I find that the rate of indel formation between human and rats is approximately 2*0.073 indels/substitution. I also find that the current, geometric, methods for gap models can give erroneous evolutionary inferences about the indel process.