[alignment], [sequence-alignment], [read-alignment] and [msa] tags

Question

Let's decide here on the status for different tags related to alignment.

There are two groups of questions.

• One is related to pairwise and multiple sequence alignments and algorithms. This field is definitely old and it was one of the keystones in the bioinformatics development. You get question like alignment tools supporting codons, RNA alignments editing or full genome alignments.

• Another one is related to read alignment/mapping. We mostly get mapping software related questions like obtaining uniquely mapped reads, BWA-MEM scoring or building genome index.

Of course read alignment is a series sequence alignment steps. In practice, however, most of the current [alignment] questions are specific only to read mapping software, and has nothing to do with sequence alignment methods.

I think we need to think about tags pragmatically, it is a way for people to follow what they are interested in, and clearly these two areas are of interest for different people. In that line, it is worth noting that meta-tags are considered useless and should be avoided.

• If the tag can’t work as the only tag on a question, it’s probably a meta-tag
• If the tag commonly means different things to different people, it’s probably a meta-tag

These are the results from our quite diverse lab, where some people are working with RNA-seq data. Clearly just [alignment] is ambiguous.

In the current form [alignment] is a meta-tag.

Meta-tagging is explicitly discouraged.

I see two ways of dealing with this.

Two separate tags

For the read alignment I propose using [read-alignment] tag. It's unambiguous. Using [alignment] wouldn't be fair because a) sequence alignment preceded read alignment b) some people (including me) still think about sequence alignment by default c) if you search google scholar for "alignment", four out of five top results will be related to sequence alignment.

For the sequence alignment I propose [sequence-alignment]. I wouldn't use [alignment] for the same reason as above (it's ambiguous). [msa] is a poor choice, since msa is considered to be an extension of pair-wise alignment.

Use [alignment] only for the sequence alignment

In this case we have to remove the [alignment] tag from all the read-mapping questions, where the alignment itself is not a problem. Maybe replacing it with [read-mapping] in some cases.

What are your suggestions?

P.S. It goes without saying, that a question related to both read alignment and sequence alignment could be tagged with the both tags. E.g. "Is it practical to use Smith–Waterman algorithm inside XYZ-mapper".

Answer 1

I am folding my previous answer, EDIT and comments into this new answer.

1. sequence-alignment includes read alignment. This is both conceptually and practically correct. The most widely used read alignment format is SAM, which is an acronym of "Sequence Alignment/Map" format. i.e. SAM is a sequence alignment format. This naming was unanimously approved by the 1000g committee consisting of many top experts in the field. Having worked with various sequence alignments for many years, I simply can't accept we exclude read alignment from sequence alignment.

2. The biggest problem with sequence-alignment is that it means different things to different people. Those in phylogenetics may think it means traditional alignment exclusively. Those who have only worked with NGS data naturally think of read alignment. In fact, this tag includes both fields. Forcing this tag to only mean traditional alignment is a wrong thing we should stop populating. It will also lead to confusion to NGS people.

3. A meta tag is an excessively broad or excessively ambiguous tag. Telling excessive broadness and ambiguity is sometimes subjective. An obvious meta tag (e.g. "best practice") can be put on almost all questions without narrowing the questions down to a specific field. Currently, we have two less obvious meta tags text and reference due to excessive ambiguity. text does not represent a field in bioinformatics. It could mean text mining or text processing, two distinct fields. Similarly reference could mean reference genome (I have retagged all questions taking this meaning) or literature, again two distinct topics. Importantly, a tag is not necessarily meta if it can be divided to more specific subfields. Actually, a true meta tag often can't be subdivided because it lacks a clear scope.

4. To me, alignment is not meta because it clearly defines a subfield in bioinformatics. If we think "alignment" is meta, "read-alignment" would be meta, too, because it includes both DNA read alignment and RNA read alignment. I can't just follow read-alignment if I am only interested in RNA-seq. In addition, alignment is frequently a safe fallback tag when the asker is not familiar with the different types of alignment. I predict it would keep popping up with time even if we retagged all current questions.

5. I don't think there is a right terminology for "alignments that are not read alignment" because each other type of alignment belongs to its own field and often uses distinct algorithms (at least partly). I think at the current stage, the best we can do is to create a tag for a smaller category of "traditional alignment". I propose to use "msa" (with "multiple-sequence-alignment" being a synonym) for MSA, "rna-structure-alignment" for RNA structure-aware alignment and "full-genome-alignment" for genome-vs-genome or assembly-vs-genome alignment.

Answer 2

I agree with this. To me the word alignment will always mean aligning sequences to each other and not aligning reads to a genome. Yes, the two are not really different under the hood, but that's not the point. Tags serve two main uses:

1. They make it easier to search for topics;
2. They help experts find questions they are interested in (through marking a tag as a favorite).

Although the underlying methods involved in both multiple sequence alignments and read mapping are not fundamentally different, the tools used for each are. Also, the people interested in each and knowledgeable about each are not always the same. I, for example, am far more familiar with sequence alignment than with aligning reads.

So, given that we have two sets of not necessarily overlapping experts and two different subjects, having two tags makes sense. The tags should serve to clearly mark what a question is about. Tagging something with alignment is not really informative since we don't know what kind of alignment we are referring to. In general, I am in favor of making tags be as descriptive and unambiguous as possible.

I would therefore be in favor of using sequence-alignment for multiple sequence alignments using tools like t_coffee of MAFFT and read-alignment for tools like bwa or bowtie that align reads to a reference. We should then remove the alignment tag since it isn't enough to explain what a question is about by itself.

Answer 3

I don't see an easy distinction between sequence-alignment and read-alignment. It could perhaps be argued that if someone is mapping and aligning lots of reads to a genomic reference, then that's read-mapping, and if alignment is not being done at the same time (e.g. Kallisto, Salmon), then that's quasi-mapping or pseudo-alignment.