inst/script/fasterOrgRecipe.R

In AnnotationForge: Tools for building SQLite-based annotation data packages

## The aim here is to sketch out and then implement a strategy for
## making all the org packages quickly by calling the makeOrgPackage()
## on code processed from the FTP file downloaded by the NCBI_ftp.R
## source.

## GOALS:
## 1) I want this to be performant.  It cannot take 30 minutes to
## download the whole FTP file and split that every time (and most of
## the time is in the splitting).  I need to do the initial processing
## step once and then run the recipe on some generated stuff that
## comes out of that. - DONE

## 2) I want to pre-filter based on a. whether the tax ID is legit,
## b. whether there are records for that id and c. whether there are
## GO annotations for that organism either at NCBI OR at blast2GO.

## 3) Some examples (see below) don't work, if it's because there is
## no data- FINE.  But the code needs to be more robust AND I need to
## not have these critters in the pre-filtered set.

## 4) It might makes sense to use the makeOrgPackage code as a base
## for making the classic org packages FIRST (it's basically a good
## opportunity to refactor).



## So lets explore an example and see what is actually the slow part...
## slow step is the part where we discard data.


debug(makeOrgDbFromNCBI)
debug(AnnotationForge:::.makeBaseDBFromDLs)
debug(AnnotationForge:::.saveFiles)
debug(AnnotationForge:::.downloadData)
debug(AnnotationForge:::.getFiles)
debug(AnnotationForge:::.writeToNCBIDB)
debug(AnnotationForge:::.indexTaxIds)


## test for older stuff
library(AnnotationForge)
makeOrgPackageFromNCBI(version = "0.1",
                       author = "Some One <so@someplace.org>",
                       maintainer = "Some One <so@someplace.org>",
                       outputDir = ".",
                       tax_id = "9606",
                       genus = "Homo",
                       species = "sapiens",
                       NCBIFilesDir=".",
                       useDeprecatedStyle=TRUE)
## WORKS
debug(AnnotationForge:::prepareDataFromNCBI)

## Another test for building the new data.frames out


debug(AnnotationForge:::prepareDataFromNCBI)


debug(AnnotationForge:::.getBlast2GO)


library(AnnotationForge)

## So this is an example where there are not any refseq or genbank IDs
## (tax ID is too general)
makeOrgPackageFromNCBI(version="0.11",
                       maintainer="g <g.ye@irri.org>",
                       author="g <g.ye@irri.org>",
                       outputDir=".",
                       tax_id="4530",
                       genus="OryZa",
                       species="sativa",
                       NCBIFilesDir=".")

## Whereas this one works just fine
makeOrgPackageFromNCBI(version="0.11",
                       maintainer="g <g.ye@irri.org>",
                       author="g <g.ye@irri.org>",
                       outputDir=".",
                       tax_id="39947",
                       genus="OryZa",
                       species="sativa.japonica",
                       NCBIFilesDir=".",
                       databaseOnly=TRUE)




## The following all works:
makeOrgPackageFromNCBI(version = "0.1",
                       author = "Some One <so@someplace.org>",
                       maintainer = "Some One <so@someplace.org>",
                       outputDir = ".",
                       tax_id = "192222",
#                       genus = "Campylobacter",
#                       species = "jejuni",
                       NCBIFilesDir=".",
                       databaseOnly=TRUE)
## This example demonstrates the limitations of using a lookup table (IOW unsupported characters can end up in the package name) 

## Lets see if this will guess the genus and species here
makeOrgPackageFromNCBI(version="0.1",
                       maintainer="Pengfei Liu <liupfskygre@gmail.com>",
                       author="Pengfei Liu <liupfskygre@gmail.com>",
                       outputDir=".",
                       tax_id="1041930",
#                       genus="Methanocella",
#                       species="conradii",
                       NCBIFilesDir=".")
## And this simply fails to find a match for this species...
## There is no helping this situation since I have not thrown out any of the available scientific names.  Users will simply have to provide a name in this case.



## Barley works...
makeOrgPackageFromNCBI(version = "0.1",
                       author = "Some One <so@someplace.org>",
                       maintainer = "Some One <so@someplace.org>",
                       outputDir = ".",
                       tax_id = "112509",
                       genus = "Hordeum",
                       species = "vulgare",
                       NCBIFilesDir=".")

## Axolotl works...
makeOrgPackageFromNCBI(version = "0.1",
                       author = "Some One <so@someplace.org>",
                       maintainer = "Some One <so@someplace.org>",
                       outputDir = ".",
                       tax_id = "8296",
                       genus = "Ambystoma",
                       species = "mexicanum",
                       NCBIFilesDir=".")






###############################################################################
## Next up: code for telling which things are viable:
## Code should 1) see if NCBI.sqlite 'cache db' exists (& if not -
## make one) and then 2) use the data in there to pre-compute which
## things are viable for making into DBs.
## (Also I need to work out how to know which things I can get data
## from blast2GO for) - pre-download this too?



###############################################################################
## AND refactor code to support the "NOSCHEMA.DB" : done

## Compare DB generated to the one made by makeOrgPackage() -
## NOSCHEMA.DB is better design.

## the older ORGANISM.db schema was unecessarily complicated. - I
## should really upgrade all these things to the new stuff right
## now. - and ALSO: the new schema allows us to make compatible with
## NOSCHEMACHIP.db packages (simpler and more extensible).
## ALSO: the new schema is less complex and better organized since it
## is generated by a hueristic (instead of being bespoke).

## What I need to do here is to support the NOSCHEMA.DB schema and
## still support making the ORGANISM.DB packages via an argument to
## makeOrgPackageFromNCBI.  Old school packages should be born in a
## 'deprecated' state.  (warn people to not use them and the argument
## should also be documented in a way that discourages it's use)

















##########################################################################
## ALSO: While I am in here I need to find a better way to deal with this:
## Basically I need an exception in place for the select code and also
## a more general solution for duplicated stuff
## goids <- keys(mouse4302.db, "GO")
## xx = select(mouse4302.db, head(goids), "GOALL", "GO")
## which takes forever and returns many duplicates
## The above is a select bug that I need to fix in AnnotationDbi...










################################################################################
################################################################################
################################################################################
################################################################################
################################################################################
## THIS is a test to see if we can really build 1100 org Dbs objects!

## Separately, I have made a list of taxIDs that I should be able to quickly process into a bunch of DBs.  I aim to test that here...

## First lets just get all the viable taxIds:

## These are the IDs (prescreened) for us to process
load(system.file('extdata','viableIDs.rda', package='AnnotationForge'))
ids <- names(results)[results]

## old school table of tax Ids
load(system.file('extdata','specData.rda', package='GenomeInfoDb'))
sd <- specData[!is.na(specData[[3]]),]

## Some taxonomy IDs cannot be looked up at all - so discard
ids <- ids[ids %in% sd$tax_id]

## AND remove this one bad one that we discovered (an overly general barley ID)
ids <- ids[!(ids %in% '4513')]

## Now ids should be OK (verified that there are species etc. before)

## need to find offenders
lookup <- function(id){
    message(paste0("looking up value for: ", id))
    AnnotationForge:::lookup_organism_by_tax_id(id)
}
res <- lapply(ids,lookup)



## Then run all (1146) of these to make sure they all work.
library(AnnotationForge)

## constant variables
version <- '3.0.0'
author = "Marc Carlson"
maintainer = "Bioconductor Package Maintainer <maintainer@bioconductor.org>"
NCBIFilesDir="."

## mapply variables
taxonomyId <- as.character(unlist(lapply(res, function(x){x$tax_id})))
genus <- unlist(lapply(res, function(x){x$genus}))
species <- unlist(lapply(res, function(x){x$species}))  
## Some name cleanup:
genus <- gsub(" ", "_", genus)
genus <- gsub("/", "|", genus)
species <- gsub(" ", "_", species)
species <- gsub("/", "|", species)


## Then go
Map(makeOrgPackageFromNCBI,
    tax_id=taxonomyId,
    genus=genus,
    species=species,
     MoreArgs=list(
       version=version,
       author=author,
       maintainer=maintainer,
       NCBIFilesDir=NCBIFilesDir,
       databaseOnly=TRUE
       ))

## Map(makeOrgPackageFromNCBI,
##     tax_id=taxonomyId[1:2],
##     genus=genus[1:2],
##     species=species[1:2],
##      MoreArgs=list(
##        version=version,
##        author=author,
##        maintainer=maintainer,
##        NCBIFilesDir=NCBIFilesDir,
##        databaseOnly=TRUE
##        ))