Feature Request: Gene Clustering for Non-Redundant Gene Catalog

[fconstancias]

Dear SqueezeMeta team,

I have been using SqueezeMeta for a while and it has been very helpful, thanks!
I was wondering if you have any plans to implement gene clustering (e.g., CD-HIT, MMseqs2) for creating a non-redundant gene catalog in the pipeline. This approach has been widely used (e.g., https://metagenome-atlas.readthedocs.io/en/latest/usage/output.html#gene-catalog, https://methods-in-microbiomics.readthedocs.io/en/latest/assembly/metagenomic_workflows.html#gene-catalogs) in the field to remove redundancy / aggregate information and fasten downstream analyses (i.e., annotation and data analyses)

Curious to hear your thoughts.

[jtamames]

Hello
Yes, I wrote some code for doing this using mmseqs2. Perhaps in future releases this will be a feature
Thanks for the suggestion!
Best,
J

[fconstancias]

That's excellent.
Since the mapping information at the (redundant) contig level is needed for binning, do you think it would be possible to link mapping information at the redundant ORFs to the ORFs clusters after the clustering is done? This would avoid mapping short reads again (this time against the representative ORFs nucleotide sequence from the non redundant sequence) to quantify the non redundant gene catalogue.
Would you agree sharing the code? Happy to discuss this further if I can help.

[jtamames]

That was my idea, mapping the reads to the representative sequences of the clusters and propagate the counts to each of the individual genes in the cluster. But I had yet to solve some issues regarding these mappings. I was doing this like one year ago and I don´t quite remember the details. I will check the code and of course share it when I fix these small issues.
Best,
J

[fconstancias]

That was my idea, mapping the reads to the representative sequences of the clusters and propagate the counts to each of the individual genes in the cluster.

The other way around (propagate the redundant / contig mapping information to the clustered ORF catalogue) would demand less modification in the SQM pipeline. What do you think?

[fconstancias]

Hello, I was wondering if you had any update on this?
Thanks

[fpusan]

As a matter of fact, yes, I did some testing by clustering the NCBI nr database using DIAMOND. Results are mixed, it definitely saves time but has an effect in annotation quality, particularly at lower taxonomic ranks.

I'm attaching a PDF with the report, including commands, execution times and annotation accuracies. Let me know if you find it useful, and I may include this as part of the official documentation.

Diamond NR clustering.pdf