R/select-methods.R

Defines functions .keysDispatch .keys .columns .select .makeActiveChrList .makeKeyList .makeJoinSQL .makeAsList .makeSelectList .tableJoinSelector .missingTableInterpolator .makeTableKey .encodeSortedTableKey .getSimpleTableNames .getTableNames .reverseColAbbreviations .makeColAbbreviations .getTableColMapping

#####################################################################
## Helpers to access/process the table names and columns
.getTableColMapping <- function(x){
  conn <- dbconn(x)
  tables <- DBI::dbListTables(conn)
  tCols <- sapply(tables, DBI::dbListFields, conn=conn)
  ## right up front we are getting rid of metadata and chrominfo tables...
  tCols[names(tCols)!="metadata" & names(tCols)!="chrominfo"]
}

## used to match and generate abbreviated column names
.makeColAbbreviations <- function(x){
  tCols <- .getTableColMapping(x)
  longNames <- unique(unlist(tCols,use.names=FALSE))
  abbrev <- unique(toupper((gsub("_","",unlist(tCols,use.names=FALSE)))))
  names(abbrev) <- longNames
  abbrev
}

## For when you need to get the true table names back from the abbrev's
.reverseColAbbreviations <- function(x, cnames){
  abr <- .makeColAbbreviations(x)
  names(abr)[abr %in% cnames] 
}

## used to retrieve vector of table names that go with vector of col names
.getTableNames <- function(x, cnames){
  realColNames <- .reverseColAbbreviations(x, cnames)
  tCols <- .getTableColMapping(x)
  ## Translate all names to one unique vector.
  getTabNames <- function(name, tCols){names(tCols[grep(name, tCols)])}
  tabNames <- lapply(realColNames, getTabNames, tCols)
  names(tabNames) <- realColNames
  tabNames
}

## This function is necessarily conservative.  It will always get all the
## possible tables that may be involved since it cannot know which ones will
## be present
.getSimpleTableNames <- function(x, cnames){
  unique(unlist(.getTableNames(x, cnames)))
}

#####################################################################
## Helpers for join-type selection

## this just takes the 1 letter abrevs and makes them into a sorted string
## that can be used as a key below
.encodeSortedTableKey <- function(sTNames){
  prefSort <- c("g","t","s","e","c")  
  res <- sTNames[match(prefSort, sTNames)]
  paste(res[!is.na(res)], collapse="")
}

.makeTableKey <- function(x,cnames){
  sTNames <- substr(.getSimpleTableNames(x, cnames),1,1)
  .encodeSortedTableKey(sTNames)
}

## for unlikely table combos 
.missingTableInterpolator <- function(tName){
  tName <- switch(EXPR = tName,
                  "se" = "tse",
                  "sc" = "tsc",
                  "te" = "tse",
                  "tc" = "tsc",
                  "ge" = "gtse",
                  "gc" = "gtsc",
                  "gs" = "gts",
                  "sec" = "tsec",
                  "gsec" = "gtsec",
                  "gtce" = "gtsec",
                  "gte" = "gtse",
                  tName)
  tName
}

## real joins for likely combos
.tableJoinSelector <- function(tName){
  ## if its not one of these, then it needs to become one
  tName <- .missingTableInterpolator(tName)
  gt <- paste("(SELECT * FROM transcript LEFT JOIN gene ",
                 "ON (transcript._tx_id = gene._tx_id) )")
  gts <- paste("(SELECT * FROM transcript LEFT JOIN gene ",
                 "ON (transcript._tx_id = gene._tx_id) INNER JOIN splicing ",
                 "ON (transcript._tx_id = splicing._tx_id) )")
  gtse <- paste("(SELECT * FROM transcript LEFT JOIN gene ",
                 "ON (transcript._tx_id = gene._tx_id) INNER JOIN splicing ",
                 "ON (transcript._tx_id = splicing._tx_id) ",
                 "INNER JOIN exon ON (splicing._exon_id = exon._exon_id) )")
  gtsc <- paste("(SELECT * FROM transcript LEFT JOIN gene ",
                 "ON (transcript._tx_id = gene._tx_id) INNER JOIN splicing ",
                 "ON (transcript._tx_id = splicing._tx_id) ",
                 "LEFT JOIN cds ON (splicing._cds_id = cds._cds_id) )")
  gtsec <- paste("(SELECT * FROM transcript LEFT JOIN gene ",
                 "ON (transcript._tx_id = gene._tx_id) INNER JOIN splicing ",
                 "ON (transcript._tx_id = splicing._tx_id) ",
                 "INNER JOIN exon ON (splicing._exon_id = exon._exon_id) ",
                 "LEFT JOIN cds ON (splicing._cds_id = cds._cds_id) )")
  ts <- paste("(SELECT * FROM transcript INNER JOIN splicing ",
              "ON (transcript._tx_id = splicing._tx_id) )")
  tse <- paste("(SELECT * FROM transcript INNER JOIN splicing ",
               "ON (transcript._tx_id = splicing._tx_id) ",
               "INNER JOIN exon ON (splicing._exon_id = exon._exon_id) )")
  tsc <- paste("(SELECT * FROM transcript INNER JOIN splicing ",
               "ON (transcript._tx_id = splicing._tx_id) ",
               "LEFT JOIN cds ON (splicing._cds_id = cds._cds_id) )")
  tsec <- paste("(SELECT * FROM transcript INNER JOIN splicing ",
                "ON (transcript._tx_id = splicing._tx_id) ",
                "INNER JOIN exon ON (splicing._exon_id = exon._exon_id) ",
                "LEFT JOIN cds ON (splicing._cds_id = cds._cds_id) )")
  
  sql <- switch(EXPR = tName,
                "g" = gt, ## becomes gt b/c we always need chr info
                "t" = "transcript", ##OK
                "s" = tse, ## becomes tse b/c we always require chr info
                "e" = "exon", ## OK thanks to exception in .makeActiveChrList()
                "c" = "cds", ## OK thanks to exception in .makeActiveChrList()
                "gt" = gt,
                "gts" = gts,
                "gtse" = gtse,
                "gtsc" = gtsc,
                "gtsec" = gtsec,
                "gse" = gtse,
                "gsc" = gtsc,
                "tse" = tse,
                "tsc" = tsc,
                "tsec" = tsec,
                stop(paste("No query for this combination of tables.",
                           "Please add",tName,"to the interpolator")))
  sql
}


#####################################################################
## Helpers to generate SQL statements for select()

## g.gene_id, s.exon_rank
## For some cols, they will occur in more than one table within the join,
## in that case, we just grab the 1st one.
.makeSelectList <- function(x, cnames, abbrev=TRUE){
    tNames <- .getTableNames(x, cnames)
    ## Here is where we only grab the 1st one...
    tNames <- lapply(tNames,function(x){x[1]})
    ## then continue on...  
    tabAbbrevs <- substr(unlist(tNames),1,1)
    names(tabAbbrevs) <- rep(names(tNames),elementNROWS(tNames))   
  if(abbrev==TRUE){
    paste(paste0(tabAbbrevs, ".", names(tabAbbrevs)), collapse=", ")
  }else{
    paste(names(tabAbbrevs), collapse=", ")
  }
}

## genes AS g, splicing AS s etc.
.makeAsList <- function(x, cnames){
  simpTNames <- .getSimpleTableNames(x, cnames)
  paste(simpTNames, "AS", substr(simpTNames,1,1), collapse=", ")
}

## WHERE g._tx_id = s._tx_id etc.
.makeJoinSQL <- function(x, cnames){
  tKey <- .makeTableKey(x,cnames)
  .tableJoinSelector(tKey)  
}

.makeKeyList <- function(x, keys, keytype, abbrev=TRUE){
  #colType <- .reverseColAbbreviations(x, keytype)
  colType <- .makeSelectList(x, keytype, abbrev)
  keys <- paste(paste0("'", keys, "'"), collapse=",")
  paste(colType, "IN (", keys,")")
}

## helper for generating where clause based on activeSeqs
.makeActiveChrList <- function(x, tKey){
  sqlCol <- switch(tKey,
                   "e" = "exon.exon_chrom",
                   "c" = "cds.cds_chrom",
                   "transcript.tx_chrom")
  chrStrings <- names(.isActiveSeq(x))[.isActiveSeq(x)]
  chrs <- paste(paste0("'", chrStrings, "'"), collapse=",")
  paste(sqlCol, "IN (", chrs,")")
}

.select <- function(x, keys, columns, keytype, ...){
  extraArgs <- list(...)
  if(missing(keys)){stop("'keys' must be a character vector")}
  if(missing(columns)){stop("'columns' must be a character vector")}
  ## Some argument checking
  if('skipValidKeysTest' %in% names(extraArgs)){
      skipValidKeysTest<-extraArgs[["skipValidKeysTest"]]}else{
          skipValidKeysTest<-FALSE}
          testSelectArgs(x, keys=keys, cols=columns, keytype=keytype,
                         skipValidKeysTest=skipValidKeysTest)
 
  ## 1st we check the keytype to see if it is valid:
  if(is.na(keys(x, keytype)[1]) & length(keys(x, keytype))==1){ 
    stop(paste("There do not appear to be any keys",
               "for the keytype you have specified."))
    }

  ## we used to add TXID to cnames, which forces splicing to always be included
  ## Splicing is a almost always needed, but almost never requested.
  ## cnames <- unique(c(columns, "TXID", keytype))
  ## 
  cnames <- unique(c(keytype, columns))
  tKey <- .makeTableKey(x,cnames)
  ## message(paste("keytype generated:",tKey))

  ## the following just gets the major join and then modifies it ONLY if the
  ## keytype is a GENEID
  majorJoin <- .makeJoinSQL(x, cnames)
  if(keytype=="GENEID"){
    majorJoin <- sub("FROM transcript LEFT JOIN gene",
                     "FROM transcript INNER JOIN gene",majorJoin)
  }

  ## This is where/how we respect isActiveSeq()
  if(FALSE %in% .isActiveSeq(x)){ 
  ## Then we have to append a where clause to majorJoin
    if(tKey %in% c("t","e","c")){
      majorJoin <- paste("( SELECT * FROM", majorJoin,
                         paste("WHERE", .makeActiveChrList(x, tKey)), ")")
    }else{
      majorJoin <- sub(") )",
                       paste(") WHERE", .makeActiveChrList(x, tKey), ")")
                       ,majorJoin)
    }
  }## otherwise we can just leave it alone.


  if(length(keys) <= 1000){  ##if more than about this, prolly faster to get all
    sql <- paste("SELECT DISTINCT",
                 .makeSelectList(x, cnames, abbrev=FALSE),
                 "FROM",
                 majorJoin,
                 "WHERE",
                 .makeKeyList(x, keys, keytype, abbrev=FALSE))  
  }else{
          sql <- paste("SELECT DISTINCT",
                 .makeSelectList(x, cnames, abbrev=FALSE),
                 "FROM",
                 majorJoin)
  }
  
  res <- AnnotationDbi:::dbQuery(dbconn(x), sql)

  if(length(keys) > 1000){ ##Then drop the extras now(in event there are some)
      ktColId <- .reverseColAbbreviations(x, keytype)
      res <-  res[res[[ktColId]] %in% keys,,drop=FALSE]
  }
  
  
  ## Then drop any columns that were not explicitely requested but that may have
  ## been appended to make a joind (like TXID)
  res <- res[,.reverseColAbbreviations(x,cnames),drop=FALSE]

  ## Then sort rows and columns and drop the filtered rows etc. using resort_base
  ## from AnnotationDbi
  joinType <- .reverseColAbbreviations(x, keytype)
  if(dim(res)[1]>0){
    res <- resort_base(res, keys, joinType,
                       .reverseColAbbreviations(x,cnames))
  }
  
#  ## Then I need to filter out rows of NAs
#  res <- res[!apply(is.na(res),1,all),,drop=FALSE]
#  ## always reset rownames after removing rows
#  rownames(res) <- NULL
  
  ## Then put the user preferred headers onto the table
  fcNames <- .makeColAbbreviations(x)
  colnames(res) <- fcNames[match(colnames(res), names(fcNames))]
  res
}


## ##  Remove this select warning function after 2.13 has released
## .selectWarnTxDb <- function(x, keys, columns, keytype, ...){
    
##     ## remove condition after 2.13
##     extraArgs <- list(...)
##     if("cols" %in% names(extraArgs)){
##         ## warn the user about the old argument
##         AnnotationDbi:::.colsArgumentWarning()
##         ## ## then call it using cols in place of columns    
##         ## if(missing(keytype)){
##         ##     .select(x, keys, columns=extraArgs[["cols"]], keytype = columns)
##         ## }else{
##         ##     .select(x, keys, columns=extraArgs[["cols"]], keytype = keytype)
##         ## }        
##     }else{
##         .select(x, keys, columns, keytype)
##     }
## }


setMethod("select", "TxDb",
    function(x, keys, columns, keytype, ...) {
##          .selectWarnTxDb(x, keys, columns, keytype, ...)
           .select(x, keys, columns, keytype, ...)
        }
)




#####################################################################
## method for columns()
.columns <- function(x){
  res <- .makeColAbbreviations(x)
  names(res) <- NULL
  res
}


setMethod("columns", "TxDb",
    function(x) .columns(x)
)



#####################################################################
## method for keys()
## PHEW, There are a lot of types.  But names are not always available so...
.keys <- function(x, keytype){
    testForValidKeytype(x, keytype)
    res <- switch(EXPR = keytype,
             "GENEID" = AnnotationDbi:::dbQuery(dbconn(x),
               "SELECT DISTINCT gene_id FROM gene", 1L),
             "TXID" =  AnnotationDbi:::dbQuery(dbconn(x),
               "SELECT DISTINCT _tx_id FROM transcript", 1L),
             "TXNAME" =  AnnotationDbi:::dbQuery(dbconn(x),
               "SELECT DISTINCT tx_name FROM transcript", 1L),
             "EXONID" =  AnnotationDbi:::dbQuery(dbconn(x),
               "SELECT DISTINCT _exon_id FROM exon", 1L),
             "EXONNAME" =  AnnotationDbi:::dbQuery(dbconn(x),
               "SELECT DISTINCT exon_name FROM exon", 1L),
             "CDSID" =  AnnotationDbi:::dbQuery(dbconn(x),
               "SELECT DISTINCT _cds_id FROM cds", 1L),                  
             "CDSNAME" =  AnnotationDbi:::dbQuery(dbconn(x),
               "SELECT DISTINCT cds_name FROM cds", 1L),                  
                  stop(paste(keytype, "is not a supported keytype.",
                             " Please use the keytypes",
                             "method to identify viable keytypes")))
    as.character(res)
}

.keysDispatch <- function(x, keytype, ...){
    if (missing(keytype)) keytype <- "GENEID"
    AnnotationDbi:::smartKeys(x=x, keytype=keytype, ..., FUN=.keys)
}
## Get the list of possible keys, for a given keytype
setMethod("keys", "TxDb",.keysDispatch)


## Examples to test:
## library(TxDb.Hsapiens.UCSC.hg19.knownGene); txdb=TxDb.Hsapiens.UCSC.hg19.knownGene
## debug(AnnotationDbi:::smartKeys)
## debug(GenomicFeatures:::tmpFun)
## debug(GenomicFeatures:::.select)
## head(keys(txdb))
## head(keys(txdb, keytype="TXNAME"))
## head(keys(txdb, keytype="TXNAME", pattern=".2$"))

## <BOOM - problem when I do column selections...
## BOTH TXID AND GENEID are keytypes AND cols so this should work...

## head(keys(txdb, keytype="TXID", column="GENEID"))

## Good grief it works!
## head(keys(txdb, keytype="GENEID", pattern=".2$", column="TXNAME"))
## select(txdb, keys = head(keys(txdb, keytype="GENEID", pattern=".2$", column="TXNAME")), cols=c("GENEID","TXNAME"), keytype="GENEID")
## select(txdb, keys = head(keys(txdb, keytype="GENEID", pattern=".2$", column="TXNAME", fuzzy=TRUE)), cols=c("GENEID","TXNAME"), keytype="GENEID")



#####################################################################
## method for keytypes()
## Seven! keytypes for TxDb
setMethod("keytypes", "TxDb",
    function(x) return(c("GENEID","TXID","TXNAME","EXONID","EXONNAME","CDSID","CDSNAME"))
)






##   library(TxDb.Hsapiens.UCSC.hg19.knownGene); x <- txdb <- TxDb.Hsapiens.UCSC.hg19.knownGene;

##   cols = c("GENEID"); keys = head(keys(x, "GENEID")); foo = select(x, keys, cols = cols, keytype="GENEID");head(foo)

##   cols = c("TXID"); keys = head(keys(x, "TXID")); foo = select(x, keys, cols = cols, keytype="TXID");head(foo)

##   cols = c("GENEID","TXID"); keys = head(keys(x, "GENEID")); foo = select(x, keys, cols = cols, keytype="GENEID");head(foo)

##   cols = c("GENEID","TXID", "EXONRANK"); keys = head(keys(x, "GENEID")); foo = select(x, keys, cols = cols, keytype="GENEID");head(foo)



##   cols = c("GENEID","TXID", "EXONRANK","CDSID"); keys = head(keys(x, "GENEID")); foo = select(x, keys, cols = cols, keytype="GENEID");head(foo)


##   cols = c("GENEID","TXID", "EXONRANK", "EXONID"); keys = head(keys(x, "GENEID")); foo = select(x, keys, cols = cols, keytype="GENEID");head(foo)


## formerly this was too slow
##   cols = c("GENEID","TXID", "EXONRANK", "EXONID", "CDSID"); keys = head(keys(x, "GENEID")); foo = select(x, keys, cols = cols, keytype="GENEID");head(foo)


## WHY is THIS a gtsec? Because our name translation is conservative.
##   cols = c("TXID", "EXONRANK", "EXONID", "CDSID"); keys = head(keys(x, "TXID")); foo = select(x, keys, cols = cols, keytype="TXID");head(foo)




##   cols = c("EXONRANK", "EXONID", "CDSID"); keys = head(keys(x, "EXONID")); foo = select(x, keys, cols = cols, keytype="EXONID");head(foo)












##  cnames = c("GENEID","TXNAME"); k = head(keys(x,keytype="GENEID")); keytype = "GENEID"; cols = c("GENEID","TXNAME")


##   keytypes(txdb)
##   head(keys(txdb, "GENEID")) 
##   head(keys(txdb, "TXID"))
##   head(keys(txdb, "EXONID"))
##   head(keys(txdb, "CDSID"))
##   head(keys(txdb, "TXNAME"))
##   head(keys(txdb, "EXONNAME"))
##   head(keys(txdb, "CDSNAME"))

##   cols(txdb)


##   AnnotationDbi:::debugSQL()  
##   k = c("foo",head(keys(txdb, "GENEID"))); foo = select(txdb, k, cols = c("GENEID","TXNAME"), keytype="GENEID"); head(foo)


## more testing:
##   cols = cols(x); k = head(keys(x,keytype="GENEID")); foo = select(txdb, k, cols = cols, keytype="GENEID");head(foo)





## A SQL bug seems to happen if I ask for a bit of everything (every table)
## and provide a key that is not an entrez gene ID

## cols = c("GENEID","TXNAME", "CDSID", "EXONSTART"); k = head(keys(txdb, "TXNAME")); foo = select(txdb, k, cols = cols, keytype="TXNAME"); head(foo)

## what is odd also is that the SQL (in a SQL session) does not seem to
## complete...


## It's not just exonstart stuff because this works:
##  cols = c("GENEID", "EXONSTART","TXNAME"); k = head(keys(txdb, "TXNAME")); foo = select(txdb, k, cols = cols, keytype="TXNAME"); head(foo)


## This works now
##  cols = c("GENEID", "CDSID","TXNAME"); k = head(keys(txdb, "TXNAME")); foo = select(txdb, k, cols = cols, keytype="TXNAME"); head(foo)




















###############################################
############ Cleared up
###############################################
## Also this gives me a strange answer (one that is not really filtered quite right) - There may be a problem with AnnotationDbi:::resort_base() - OR maybe I can just put a "DISTINCT" into the query? - and maybe I should really do both???

## cols = c("GENEID","TXNAME", "TXID"); k = head(keys(txdb, "TXID"));foo = select(txdb, k, cols = cols, keytype="TXID"); head(foo)

## Issue (used to) exist with above where the txid is NOT actually being
## filtered
## right.  The answer that comes back looks like:
##   TXID GENEID     TXNAME
## 1    1   <NA>       <NA>
## 2    2   <NA>       <NA>
## 3    3   <NA>       <NA>
## 4    4 653635 uc009vis.2
## 5    4 653635 uc009vis.2
## 6    4 653635 uc009vis.2
## ## but should look like:
##   TXID GENEID     TXNAME
## 1    1   <NA>       <NA>
## 2    2   <NA>       <NA>
## 3    3   <NA>       <NA>
## 4    4 653635 uc009vis.2
## 5    5 653635 uc009vjc.1
## 6    6 653635 uc009vjd.2


## DONE: add checks to pre-warn when a keytype has no values for it's key
## cols = c("GENEID","TXNAME", "TXID", "CDSNAME"); k = head(keys(txdb, "CDSNAME"));foo = select(txdb, k, cols = cols, keytype="CDSNAME"); head(foo)




## The following breaks because I end up with a jointype of _cds_id, but with
## NO COLLUMNS that match to it..  So my SQL code has failed to retrieve
## everything I needed...

## IOW, the res has content from the DB but not the right amount?
## cols = c("GENEID","TXNAME", "TXID"); k = head(keys(txdb, "CDSID"));foo = select(txdb, k, cols = cols, keytype="CDSID"); head(foo)


## Another example of this:
## The following wasn't working because the key column must always be part of
## the query. (fixed)

##  debug(AnnotationDbi:::resort_base); AnnotationDbi:::debugSQL();
##  cols = c("GENEID","TXNAME", "TXID", "CDSNAME"); k = head(keys(txdb, "CDSID"));foo = select(txdb, k, cols = cols, keytype="CDSID"); head(foo)

## Drop rows that contain only NAs. - This is a good cleanup generalyl, and also if you drop the column that you sorted on then you may have added a bunch of NA rows for where there was no data, and now those are just silly. - DONE






#### So what is the pattern?
#### prefix (add later)
## SELECT * FROM

#### OPTIONAL: swap in as clause code for "gene AS g" etc.
#### OPTIONAL EFFICIENCY: replace * from core clauses with string from .makeSelectList() (but after filtering out unwanted stuff of course) so that we only grab stuff that was asked for from the beginning.
#### OPTIONAL EFFICIENCY???: Can I make it smarter so that it starts with the table that has the selected keys and then builds out from there?  No that can't work, because of the outer join requirements?? - definitely want to do this one "later" since even if it can work it will be difficult to implement gracefully.

#### create 8 core-combos
#### and select appropriate one based on who is in .getSimpleTableNames()
## (SELECT * FROM gene AS g LEFT JOIN transcript AS t ON
## g._tx_id=t._tx_id
## UNION
## SELECT * FROM transcript AS t LEFT JOIN gene AS g ON
## g._tx_id=t._tx_id )

#### suffix (add later, once we have our cores)
## WHERE gene_id IN ('10772','22947')
## limit 10;


## This tests the "gse"
## library(TxDb.Hsapiens.UCSC.hg19.knownGene);x <- txdb <- TxDb.Hsapiens.UCSC.hg19.knownGene; foo = select(txdb, keys=head(keys(txdb)) , cols=c("EXONSTART","GENEID","EXONRANK"),keytype="GENEID")
## This tests whether or not I get the keytype back with select...
## foo = select(txdb, keys=head(keys(txdb)) , cols=c("EXONSTART","EXONRANK"),keytype="GENEID")












## NEW STRESS TEST (big join with A LOT OF gene_id keys
## library(TxDb.Hsapiens.UCSC.hg19.knownGene);x <- txdb <- TxDb.Hsapiens.UCSC.hg19.knownGene;  cols = c("GENEID","CDSSTART"); keys = keys(x, "GENEID"); foo = select(x, keys, cols = cols, keytype="GENEID");head(foo)




## library(TxDb.Hsapiens.UCSC.hg19.knownGene);x <- TxDb.Hsapiens.UCSC.hg19.knownGene; cols = c("GENEID","CDSSTART", "CDSCHROM");keys = head(keys(x, "GENEID"));
## select(x, keys, cols = cols, keytype="GENEID");
## .isActiveSeq(x)[seqlevels(x)] <- FALSE;
## .isActiveSeq(x) <- c("chr1"=TRUE)
## select(x, keys, cols = cols, keytype="GENEID");

##  debug(GenomicFeatures:::.select)


## cols = c("TXNAME"   ,  "TXCHROM" ,   "TXSTRAND")
## k = head(keys(x,keytype="TXNAME"))
## select(x, k, cols = cols, keytype="TXNAME");



##
## So I think I have this fixed now.  But I still need to add unit tests.





###################3##########################################################
## New bug.  Can reproduce it like this:
## FIXED
## library(TxDb.Hsapiens.UCSC.hg19.knownGene); x <- TxDb.Hsapiens.UCSC.hg19.knownGene; geneGR <- exonsBy(x, "gene"); keys <- names(geneGR); kt = "GENEID"; cols = "GENEID"; debug(GenomicFeatures:::.select)
## GenomicFeatures:::.select(x, keys, columns=cols, keytype=kt)
jmacdon/GenomicFeatures documentation built on Jan. 2, 2022, 7:40 a.m.