search

bluenote-1577 / sylph

ultrafast taxonomic profiling and genome querying for metagenomic samples by abundance-corrected minhash.
MIT License
185 stars 6 forks source link

[FEATURE REQUESTS] - post here for suggestions/feature requests #6

Open bluenote-1577 opened 10 months ago

bluenote-1577 commented 10 months ago

Feature requests

Purpose: this is a place to easily log suggestions/feature requests. E.g:

Give a rationale and provide concise/clear instructions if possible. Opinions are welcome too.

You're welcome to email me or open another issue. This thread is to aggregate suggestions without the hassle of opening another issue.

Current feature requests

Here are some current feature requests.

Originally posted by @jolespin in https://github.com/bluenote-1577/skani/issues/23#issuecomment-1858723727

@fplaza https://github.com/bluenote-1577/sylph/issues/6#issuecomment-1859073351

fplazaonate commented 10 months ago

Hi Jim, Here is a suggestion: Save read length while sketching so the user does not have to provide it to compute true coverage.

astrovsky01 commented 7 months ago

Is there/could there be a method to output unassigned reads into a file as an output?

bluenote-1577 commented 7 months ago

Hi @astrovsky01,

This is unfortunately not possible due to the way sylph works. It doesn't classify each read. It operates on the ensemble of reads. This means it can not output unassigned reads, only estimate the percentage of.

jolespin commented 3 months ago

@bluenote-1577 I'm not sure how the backend algorithm works but is it possible to add Align_fraction_ref to Sylph output similar to Skani? Would be useful to know how much of the genomes being profiled are covered.

bluenote-1577 commented 3 months ago

@jolespin Hi Josh, unfortunately this isn't possible. This is because skani actually tries to get a pseudo-ish alignment, by sylph doesn't do anything like that. I agree it would be very nice if it were possible though...

jolespin commented 3 months ago

Ok that's good to know! Would finding the overlap in kmers do the trick or is it way more complicated than that?

jolespin commented 3 months ago

Also one more question, does sylph allow for outputting abundance instead of relative abundance?

bluenote-1577 commented 3 months ago

@jolespin sorry for the late response:

1) There may be something that could be done for pseudo-alignment overlapping k-mers ... but it's a very nontrivial algorithmic thing :)

2) What do you mean by abundance instead of relative abundance? Sylph outputs coverage (Est_cov) if that's what could be helpful

jolespin commented 3 months ago

There may be something that could be done for pseudo-alignment overlapping k-mers ... but it's a very nontrivial algorithmic thing :)

I can imagine that is quite complicated. Themisto just popped up on my radar so I'm going to give this a try soon.

What do you mean by abundance instead of relative abundance? Sylph outputs coverage (Est_cov) if that's what could be helpful

Is the Est_cov what you use before normalizing the Taxonomic abundance?

Taxonomic_abundance: normalized taxonomic abundance as a percentage. Coverage-normalized - same as MetaPhlAn abundance https://github.com/bluenote-1577/sylph/wiki/Output-format

I'm mostly curious on how some of my compositionally valid network analysis (https://github.com/jolespin/ensemble_networkx) differs between coverage normalized and unnormalized data but definitely not a critical assessment. Just a bit of curiosity.

jolespin commented 1 month ago

This is unfortunately not possible due to the way sylph works. It doesn't classify each read. It operates on the ensemble of reads. This means it can not output unassigned reads, only estimate the percentage of.

I'm looking at the docs now and not sure which field indicates the % of reads/k-mers not aligned/overlapping(sorry if that's the wrong term) with the k-mers in the database/sketch. If this isn't currently available, would it be possible to add this metric? It would greatly benefit my workflow when determining whether or not I want to assemble/bin genomes from a metagenomic assembly.

bluenote-1577 commented 1 month ago

@jolespin

You'll notice that the "Sequence abundance" column doesn't sum to 100% if the -u option is specified (unless sylph determines that 100% of your reads are classified at species level). Ths sum of this column indicates the % of reads classified at species level. That is, the sequence abundance is scaled by the % of classified reads. So

So without -u, you'll get

Species 1 50% Species 2 50%

But with -u, if 10% of the reads come from species-level detected genomes, you'll get

Species 1 5% Species 2 5%

let me know if that makes sense. I didn't want to add a new column because it doesn't really make sense... but this is a bit non-obvious to see

jolespin commented 1 month ago

Excellent! Love that functionality. So essentially, I can just run -u and then w/ Pandas do something like X["unclassified"] = 100 - X.sum(axis=1) if I want the % of unclassified reads in the table. Alternatively, I can do (X/X.sum(axis=1).values.reshape(-1,1)) * 100 before I add the unclassified column to get the relative abundance of the taxa.

bluenote-1577 commented 1 month ago

@jolespin yes exactly. BTW, I discovered the "discussion" feature in github. I think we can migrate the questions there, as well as suggestions, perhaps keeping this thread for specific feature requests.