eggNOG: DNA-DNA or protein-protein orthologue identification?

Question

I'm trying to work out is how reliable/accurate my eggNOG metagenomic outputs are. Critical for this is how its identifying orthologues.

Background eggNOG is commonly used in metagenomics gene identification. Take a sample, randomly sequence all the DNA therein, then "annotate" via eggNOG. It identifies the putative gene function against lots of databases e.g. NCBI, KEGG, Brite, Go ...

eggNOG uses is own database for gene identification via 'seed orthology' each seed has its own KEGG ID.

Question Does the comparison between the query (metagenomics data) and the seeds take place using DNA-DNA comparisons? Alternatively, is it performing protein-protein comparisons? Algorithmically, by the easiest is DNA-DNA, however DNA-DNA would result in limitations of interpretation ('cause of complex evolutionary stuff). It could be using an optimised a protein database instead and returning the output as DNA.

Answer 1

I can answer this now. I've been 'under the hood' of eggNOG and its using "hmm" (hidden Markov model) very likely, almost certain, the Eddie (sp?) group's hmmer to cluster proteins associated. It represents the cluster members strictly via seed orthologue code, which then are aligned to form a 'raw' and 'trimmed' alignment. Each alignment is assigned a code which corresponds to its know function. This is a database of alignments. Once it gets a protein - protein hit against one of its orthologue (I suspect it will be using a nested system) it will then assign the query to that alignment - each of which is assigned to a protein function.

It representative database not an all encompassing database, i.e. sequences are being omitted.

Taxonomically it is simply declaring a the familial name. There will be an internal database mapping the seed orthologue onto the species, but that is not being publicly disclosed.

I make a better explanation than that if anyone wants it.

---

There is some description in the original paper