--- title: "AnnotationHub Recipes" author: "Marc Carlson, Sonali Arora" date: "`r BiocStyle::doc_date()`" package: "`r BiocStyle::pkg_ver('BiocAnnotRes2015')`" abstract: > AnnotationHub Recipes vignette: > %\VignetteIndexEntry{AnnotationHub Recipes} %\VignetteEncoding{UTF-8} %\VignetteEngine{knitr::rmarkdown} output: BiocStyle::html_document: toc: true --- ```{r style, echo = FALSE, results = 'asis'} BiocStyle::markdown() ``` ```{r, echo=FALSE, results="hide"} library(knitr) opts_chunk$set(error=FALSE) ``` --- Authors: Marc RJ Carlson and Sonali Arora Key Words: annotation, next-generation sequencing, R, Bioconductor --- ## Overview of the process If you are reading this it is (hopefully) because you intend to write some code that will allow the processing of online resources into R objects that are to be made available via that the AnnotationHub package. In order to do this you will have to do four basic steps (outlined below). These steps will have you writing two functions and then calling a third function to do some automatic set up for you. The 1st function will contain instructions on how to process data that is stored online into metadata for describing your new R resources for the AnnotationHub. And the 2nd function is for describing how to take these online resources and transform them into an R object that is useful to end users. ## Setup It should go without saying that this vignette is intended for users who are comfortable with R. And in order to follow the instuctions in this vignette, you will need to install the AnnotationHubData package. This package is not meant to be used by most people, and in fact it's not really intended to be anything other than a support package. So it's not exposed via biocLite(). So to get it you will need to use svn to check it out from the following location: ```{} https://hedgehog.fhcrc.org/bioconductor/trunk/madman/Rpacks/AnnotationHubData ``` Once you have that checked out, you will need to use R CMD INSTALL to install the package from source. ## Introducing AnnotationHubMetadata Objects The AnnotationHubData package is a complementary package to the AnnotationHub package that provides a place where we can store code that processes online resources into R objects suitable for access through the AnnotationHub package. But before you can understand the requirements for this package it is important that you 1st learn about the objects that are used as intermediaries between the hub and its web based repository behind the scenes. That means that you need to know about AnnotationHubMetadata objects. These objects store the metadata that describes an online resource. And if you want to see a set of online resources added to the repository and maintained, then it will be necessary to become familiar with the AnnotationHubMetadata constructor. For each online resource that you want to process into the AnnotationHub, you will have to be able to construct an AnnotationHubMetadata object that describes it in detail and that specifies where the recipe function lives. The steps involved include writing a recipe which adds files to AnnotationHub and can be summarized briefly as : - Writing a function which takes the metadata about the resource and processes them into AnnotationHubMetadata objects - Optional step : Write an additional function specifying how the files need to pre-processed. The data from these files is transformed into an R object that is useful to end users. - Optional step: Write a function specifying how the files need to be post-processed once downloaded to a user's local cache. ## Step 1: Writing your AnnotationHubMetadata generating function The following example function takes files from the latest release of inparanoid and processes them into AnnotationHubMetadata objects using Map. The 1st function you need to provide is one that processes some online resources into AnnotationHubMetadata objects. This function MUST return a list of AnnotationHubMetadata objects. It can rely on other helper functions that you define, but ultimately it (and it's helpers) need to contain all of the instructions needed to find resources and process those resources into AnnotationHubMetadata objects. The calling of the Map function is really the important part of this function, as it shows the function creating a series of AnnotationHubMetadata objects. Prior to that, the function was just calling out to other helper functions in order to process the metadata so that it could be passed to the AnnotationHubMetadata constructor using Map. Notice how one of the fields specified by this function is the Recipe, which indicates both the name and location of the recipe function. We expect most people will want to submit their recipe to the same package as they are submitting their metadata processing function. ```{r, exampleInpProcessing} makeinparanoid8ToAHMs <- function(currentMetadata){ baseUrl <- 'http://inparanoid.sbc.su.se/download/current/Orthologs_other_formats' ## Make list of metadata in a helper function meta <- .inparanoidMetadataFromUrl(baseUrl) ## then make AnnotationHubMetadata objects. Map(AnnotationHubMetadata, Description=meta$description, Genome=meta$genome, SourceFile=meta$sourceFile, SourceUrl=meta$sourceUrl, SourceVersion=meta$sourceVersion, Species=meta$species, TaxonomyId=meta$taxonomyId, Title=meta$title, RDataPath=meta$rDataPath, MoreArgs=list( Coordinate_1_based = TRUE, DataProvider = baseUrl, Maintainer = "Marc Carlson ", RDataClass = "SQLiteFile", RDataDateAdded = Sys.time(), RDataVersion = "0.0.1", Recipe = "AnnotationHubData:::inparanoid8ToDbsRecipe", Tags = c("Inparanoid", "Gene", "Homology", "Annotation"))) } ``` Now before we move on on to step two here is a listing of the different arguments that the AnntotationHubMetadata object can take and what class is expected for each of them: ```{} AnnotationHubRoot: 'character(1)' Absolute path to directory structure containing resources to be added to AnnotationHub SourceUrl: 'character()' URL where resource(s) can be found SourceType: 'character()' which indicates what kind of resource was initially processed. The preference is to name the type of resource if it's a single file type and to name where the resources came from if it is a compound resource. So Typical answers would be like: 'BED','FASTA' or 'Inparanoid' etc. SourceVersion: 'character(1)' Version of original file SourceLastModifiedDate: 'POSIXct()' The date when the source was last modified. Leaving this blank should allow the values to be retrieved for you (if your sourceURL is valid). SourceMd5: 'character()' md5 hash of original file SourceSize: 'numeric(1)' Number of bytes in original file DataProvider: 'character(1)' Where did this resource come from? Title: 'character(1)' Title for this resource Description: 'character(1)' Description of the resource Species: 'character(1)' Species name TaxonomyId: 'character(1)' NCBI code Genome: 'character(1)' Name of genome build Tags: 'character()' Free-form tags Recipe: 'character(1)' Name of recipe function RDataClass: 'character(1)' Class of derived object (e.g. 'GRanges') RDataDateAdded: 'POSIXct()' Date added to AnnotationHub. Used to determine snapshots. RDataPath: 'character(1)' file path to serialized form Maintainer: 'character(1)' Maintainer name and email address, 'A Maintainer ' BiocVersion: 'character(1)' Under which resource was built Coordinate_1_based: 'logical(1)' Do coordinates start with 1 or 0? DispatchClass: 'character(1)' string used to indicate which code should be called by the client when the resource is downloaded. This is often the same as the RDataClass. But it is allowed to be a different value so that the client can do something different internally if required. Location_Prefix: 'character(1)' This was added for resources where the metadata only is stored and the resource itself comes from a third party web site. The location prefix says the base path where the resource is coming from, and the default value will be from our own site. Notes: 'character()' Notes about the resource. ``` ## Step 2: Function for pre-processing the File (Recipe) The 2nd kind of function you need to write is called a recipe function. It always must take an single AnnotationHubMetadata object as an argument. The job of a recipe function is to use the metadata in an AnnotationHubMetadata object to produce an R object or data file that will be retrievable from the AnnotationHub service later on. Below is a recipe function that calls some helper functions to generate an inparanoid database object from the metadata stored in it's AnnotationHubMetadata object. ```{r, exampleRecipe} inparanoid8ToDbsRecipe <- function(ahm){ require(AnnotationForge) inputFiles <- metadata(ahm)$SourceFile dbname <- makeInpDb(dir=file.path(inputFiles,""), dataDir=tempdir()) db <- loadDb(file=dbname) outputPath <- file.path(metadata(ahm)$AnnotationHubRoot, metadata(ahm)$RDataPath) saveDb(db, file=outputPath) outputFile(ahm) } ``` ## Note for step 1 and step 2 While writing this function, care has to be taken for a couple of fields: Case 1 - If the file just needs to be downloaded and only post-processed in users local cache then 1) SourceUrls = Location_Prefix + RDataPath 2) Recipe = NA_character_ Example - ```{} SourceUrls="http://hgdownload.cse.ucsc.edu/goldenPath/hg38/liftOver/hg38ToRn5.over.chain.gz", RDataPath="goldenPath/hg38/liftOver/hg38ToRn5.over.chain.gz" , Location_Prefix = "http://hgdownload.cse.ucsc.edu/", ``` Case 2 - If the recipe needs to retrieve a file from an external website, pre-process it, store this pre-processed file at our amazon location and always render the pre-processed file ( not the original file) to the user then 1) SourceUrls should merely document the original location of the untouched file 2) Location_Prefix + RDataPath should be equal to the file path on the amazon machine where all pre-processed files are stored. 3) Recipe = helper function which tells us how to pre-process the original file Example - ```{} SourceUrls="http://hgdownload.cse.ucsc.edu/goldenPath/hg38/liftOver/hg38ToRn5.over.chain.gz", Location_Prefix = "http://s3.amazonaws.com/annotationhub/", RDataPath="chainfile/dummy.Rda" ``` If this seems confusing, please note how in both of these cases, the sourceUrl needs to reflect the location that the resource will actually come from once when the client is in use. ## Step 3: Function for Post-processing a File in User's cache. One can post-process the file when it is instantiated into AnnotationHub from the user's cache. An example, would be a BED file is downloaded to the user's cache, and we want AnnotationHub to read it as a `GRanges` using `rtrackler::import` Then along with your recipe, one would write a class to be included inside AnnotationHub as shown below- ```{r eval=FALSE} setClass("BEDFileResource", contains="AnnotationHubResource") setMethod(".get1", "BEDFileResource", function(x, ...) { .require("rtracklayer") yy <- .hub(x) dat <- rtracklayer::BEDFile(cache(yy)) rtracklayer::import(dat, format="bed", genome=yy$genome, ...) }) ``` If you need to do this with a set of files that you are crafting a recipe for, you will need to coordinate with us so that we can patch the appropriate supporting code into the client. Alternatively, you can make sure to set the RDataClass to an existing value (one that we already have a method for). ## Step 4: Test your functions and then contact us when they work So at this point you should make sure that the AnnotationHubMetadata generating function produces a list of AnnotationHubMetadata objects and that your recipe produces a path to a file that is generated in the way that you expect it to. Once this happens you should contact us about running your recipe so that your data can actually be put into the hub. ## Session Information ```{r, SessionInfo, echo=FALSE} sessionInfo() ```