2
$\begingroup$

I was reading the AlphaFold paper and had difficulty with a couple of terms introduced in the main text of the paper. I asked ChatGPT what these were but I'm not sure that it's accurate.

I had a hard time finding the definitions of an MSA (multiple sequence alignment) cluster and MSA depth. GPT-4 explains that an MSA cluster is a subset of sequences from an overall MSA that are closely related to each other, which seems to be simply the results from clustering analysis on MSA. GPT-4 also defines the MSA depth as the number of sequences in MSA. However, I want to make sure that these definitions are correct and are not just hallucinations from GPT-4.

$\endgroup$
6
  • 2
    $\begingroup$ Please finish reading the Tour and the Help on asking questions. Don't use non-standard abbreviations without definition. I have run MSA for years and never heard of GPT-4. Sure, I may be old and out of touch, but very few list members will have heard of it either. Say what it is and provide a link. If you can't find standard definitions for terms, say on Wikipedia, it is likely that there won't be any. Something like "cluster" and "depth" would seem so general that I doubt whether there can be any "correct" definition of them. What are you trying to do that you care? $\endgroup$
    – David
    Commented Sep 18, 2023 at 10:38
  • $\begingroup$ Where did these terms come from? Are you just reading GPT-4 outputs, or did you find the terms and ask GPT-4 to explain them? If the former, I wouldn't trust the chatbot to understand differences between local/global alignment (see other answer), and thus it is probably mixing a lot of apples and oranges in a nonhelpful way. If the latter, it would be much more useful to see the original source, which we might be able to explain. $\endgroup$ Commented Sep 18, 2023 at 16:47
  • 1
    $\begingroup$ Hi @Maximilian Press, thanks for the reply and sorry for not providing more context. I was using AlphaFold and one of the parameters was for specifying the number of MSA clusters, I then asked GPT-4 for its definition. AlphaFold was introduced in this paper for protein structure prediction and is also where I saw the term "MSA depth": nature.com/articles/s41586-021-03819-2 $\endgroup$
    – Jack
    Commented Sep 18, 2023 at 18:55
  • $\begingroup$ @Jack: thanks for following up, I edited your question and provided an answer below. $\endgroup$ Commented Sep 19, 2023 at 17:21
  • 1
    $\begingroup$ I'd really recommend staying away from ChatGPT and similar for learning for now; these tools are really great at creating plausible text, but they can't cite sources so you can't follow up, and they have no concept of what is "correct" or not, they are just using a statistical model to predict which words should come next in a sequence. That means they can sometimes do a good job of summarizing content that is repeated over and over again in their training set, but also that sometimes it gets mixed up and confuses concepts, misses a negative, or confidently states something absolutely wrong. $\endgroup$
    – Bryan Krause
    Commented Sep 19, 2023 at 17:23

1 Answer 1

1
$\begingroup$

AlphaFold defines these measures in the supplementary information of the paper, as referenced in the main text.

Generally speaking, a lot of information for these highly condensed papers goes into the supplementary information, and that's the first place to look for more explanation.

MSA depth

Figure S10 legend:

MSA depth is computed by counting the number of non-gap residues for each position in the MSA (using the Neff weighting scheme with a threshold of 80% identity measured on the region that is non-gap in either sequence) and taking the median across residues.

This is not a standard measure, just to be clear.

MSA clustering

There is a section "1.2.7 MSA clustering" in the paper supplement that describes the clustering procedure at some length.

MSA clustering is a fairly standard thing to do, but there are many possible algorithms for deriving clusters, and a large family of implementations of each algorithm. I suggest looking into the field of phylogenetic inference for other examples.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .