Feature request: Sequence similarity metric

Hi all, I was thinking that it would be useful to provide a sequence similarity metric to enable comparing protein or DNA sequences. The metric could compute all-by-all similarities or even employ some more sophisticated techniques to produce a sparse similarity matrix or the neighbourhood graph directly.

Is this something you would consider, or do you think this is out of scope?

The current way of solving it is to use some other tool to generate a distance matrix or a neighbourhood graph matrix and then embed it using UMAP precomputed or exact metric. But this is non-trivial and there are no relevant tutorials that would demonstrate this (at least not that I know of).

The issue names no files, tests, or entry points. Start by reviewing UMAP's support for precomputed or exact metrics and the stated workflow of supplying a distance matrix or neighbourhood graph. Done would require an agreed scope for sequence similarity, including whether to produce all-by-all similarities, a sparse matrix, or a neighbourhood graph, plus relevant documentation or tutorials.