kmerize with hg19 complains of "long vectors not supported yet"

Greetings,

I am running the mappable function for hg19 genome:

library(kmap)
library(BiocParallel)

# either
mappable.regions<-mappable("hg19",kmer=50,BPPARAM=MulticoreParam(workers = 1))

# or
mappable.regions<-mappable("hg19",kmer=50)

But finally I get the following error:

INFO [2018-06-01 15:09:12] Removing non-standard DNA bases
INFO [2018-06-01 15:26:38] Chopping into 50-mers
Error in .Call2("valid_Ranges", x_start, x_end, x_width, PACKAGE = "IRanges") :
long vectors not supported yet: memory.c:3451

Any idea what might has caused such an error?

[omsai]

Yeah, I need to architecturally overhaul how the storage is managed to workaround R's integer limit. This might be similar to #6

At the minute, hg19 can't be kmerized because R has a hard limit on vector maximum size as 2.147e+9 (see https://stackoverflow.com/a/21142236). As the stackoverflow answer explains, when you access vectors using the array notation, underneath the covers the indexes use integers, but hg19 is > 3 billion basepairs:

> suppressPackageStartupMessages(library(BSgenome.Hsapiens.UCSC.hg19))
> message(format(2^31-1, big.mark = ","))
2,147,483,647
> message(format(sum(as.numeric(elementNROWS(BSgenome.Hsapiens.UCSC.hg19))), big.mark = ","))
3,137,161,264
> sum(elementNROWS(BSgenome.Hsapiens.UCSC.hg19))
[1] NA
Warning message:
In sum(elementNROWS(BSgenome.Hsapiens.UCSC.hg19)) :
  integer overflow - use sum(as.numeric(.))
>

Thanks for the feedback!

[omsai]

Actually, the architectural bug about long vectors was fixed upstream. Did a bit more searching for "long vectors bioconductor" and found this issue was fixed in R-3.4 devel:
https://support.bioconductor.org/p/101439/
The workaround for R 3.4 of run useDevel(), as Aaron Lun suggests in the link above, no longer works for me; probably because the newer Bioconductor devel was released for R 3.5:

> version$version.string
[1] "R version 3.4.2 (2017-09-28)"
> BiocInstaller::useDevel()
Error: 'devel' version not available

This leaves upgrading to R-3.5:

# In R 3.5:
source("https://bioconductor.org/biocLite.R")
biocLite(c("devtools", "BSgenome.Hsapiens.UCSC.hg19"))
devtools::install_github("coregenomics/kmap", repos = BiocInstaller::biocinstallRepos())
library(kmap)
library(BiocParallel)
mappable_regions <- mappable("hg19", kmer = 50, BPPARAM = SerialParam())

I haven't tried testing this yet because my lab machine keeps running out of memory even with the single core. I'm in the process of installing R 3.5 and trying kmap on our university cluster and will let you know.

[omsai]

Fault persists with R 3.5 and running kmap on the university cluster:

INFO [2018-06-11 23:24:38] Removing non-standard DNA bases
INFO [2018-06-11 23:33:29] Chopping into 50-mers
Error in .Call2("Ranges_validate", x_start, x_end, x_width, PACKAGE = "IRanges") : 
  long vectors not supported yet: memory.c:3486
Calls: mappable ... anyStrings -> isTRUE -> validityMethod -> valid.func -> .Call2

Will post to the bioc-devel mailing list.