Hello, 

I am trying to improve the way that I am retrieving genes related to a go term, getting something closer to amigo/quickgo (are they the most up to date ?)
I have two simple ways so far: 

    AnnotationDbi::select(org.Hs.eg.db,
                              keys=go_key,
                              columns = c('SYMBOL'),
                              keytype = "GOALL")

and 

    ensembl <- biomaRt::useMart(biomart = "ensembl", dataset = "hsapiens_gene_ensembl")
    biomaRt::getBM(attributes=c('hgnc_symbol', 'ensembl_transcript_id', 'go_id', 'go_linkage_type'), filters = 'go', values = go_key, mart = ensembl)

I kind of like Annotationdbi way better as I have the db on my computer and therefore is it fast, no need to wait for internet request/response. 

My first problem is they return slightly different gene list which are also slightly different compare to amigo/quickgo.

Do you know with one is the more updated ? Also I am not sure the GOALL is retrieving all the child and child of child ... maybe I should experiment with recursive method. 

The GO.db package is built twice a year, so at best the one you have is from April. The data at the Biomart server is updated, um, more often? I presume there is some information somewhere that would tell you, or perhaps Mike Smith knows. I sort of assumed that they updated more regularly, but now that I think about it I am not sure why I would think that.

I would assume that AmiGO is the most up to date, given that it's their data.

But as you have noted, the OrgDb is the fastest way to go, and also is nice because it's constant (and hence reproducible). There is always a tension between getting the newest information and being able to reproduce results, and people have to decide for themselves what is most important.

Also, you don't have to guess at GOALL, you can just look at the help page. From ?GOALL:

 GO: GO Identifiers associated with a gene of interest

 GOALL: GO Identifiers (includes less specific terms)