Skip to content

Latest commit

 

History

History
257 lines (167 loc) · 7.81 KB

migration_notes.md

File metadata and controls

257 lines (167 loc) · 7.81 KB

Migration of Rsamtools to Rhtslib

What is the migration about?

Before this migration (i.e. for versions < 1.99.0), Rsamtools contained a copy of an old (9-10 year old?) version of the samtools and tabix C code.

The latest version of samtools (as of Feb 6, 2019) is 1.9: http://www.htslib.org/

In the recent years the samtools code has been split into samtools + htslib. Most of the code that used to be in samtools is now in htslib. The latest version of htslib is also 1.9.

Bioconductor package Rhtslib contains htslib 1.7, which is a decently recent version of htslib.

After this migration (i.e. for versions >= 1.99.0), Rsamtools no longer contains a copy of the samtools and tabix C code but compiles and links against Rhtslib instead (i.e. against htslib 1.7).

Current migration status

C/C++ code

In Rsamtools 1.99.0 all the C/C++ code from the old Rsamtools was migrated to Rhtslib, except for the code behind asBcf() and razip():

  • asBcf() (implemented in src/bcffile.c) is currently disabled but will need to be migrated if it turns out that some other package needs it (it doesn't seem to be the case though). Note that asBcf() isn't used anywhere in Rsamtools either (no example in the man page, no unit test, not used in the vignette).

  • razip() (implemented in zip_compression.c in the old Rsamtools) was removed. This is because recent samtools and htslib no longer support RAZF. From the samtools NEWS file:

    RAZF and razip are superseded by BGZF/bgzip and have been removed from samtools.

    This happened in samtools 1.0 (15 August, 2014).

Other remaining problems

  • Even though the C code behind applyPileups() was migrated (pileupbam.c), the examples in its man page and most of the tests in inst/unitTests/test_applyPileups.R are currently failing and have been disabled for now.

  • Test test_BcfFile_scanBcfHeader_remote (located in inst/unitTests/test_BcfFile.R) passes when run interactively (or via BiocGenerics:::testPackage('Rsamtools')) but, for some obscure reason, NOT in the context of R CMD check. It has been disabled for now.

Status of Bioconductor packages that depend directly on Rsamtools

As of Feb 6, 2019, 159 Bioconductor packages (154 software, 3 data-experiment, and 2 workflow packages) depend directly on Rsamtools (i.e. via their Depends, Imports, or LinkingTo field). So we've only checked manually a few of them. This was on a 64-bit Ubuntu 16.04.5 LTS laptop running R 3.6 (2018-12-04 r75758) and Bioconductor 3.9 (current devel).

All software packages with "LinkingTo: Rsamtools" were tested

  • VariantAnnotation: migrated (changes committed), passes R CMD check

  • ArrayExpressHTS: migrated (changes not committed yet, Angela Goncalves contacted but didn't answer yet), passes R CMD check

  • BitSeq: migrated (changes committed), passes R CMD check

  • DiffBind: migrated (changes committed by Gord Brown), passes R CMD check

  • h5vc: migrated (changes committed, including fixing pre-existing build ERROR currently visible on the build report), passes R CMD check

  • podkat: migrated (changes committed, including fixing inst/examples/example1.vcf.gz to make it compatible with bcftools 1.7), passes R CMD check

  • qrqc: migrated (PR created at vsbuffalo/qrqc#6 per Vince Buffalo's request), passes R CMD check

  • QuasR: migrated (PR created at fmicompbio/QuasR#9 per Michael Stadler's request), passes R CMD check

  • seqbias: migrated (patch sent to, and applied by, Daniel Jones), passes R CMD check

  • TransView: migrated (changes committed), passes R CMD check

All software packages that use applyPileups() were tested

  • AllelicImbalance: migrated (contained old BCF file (ERP000101.bcf, in extdata/ERP000101_subset/) that needed to be fixed and reindexed because it no longer worked with recent bcftools/Rhtslib) passes R CMD check

  • biovizBase: passes R CMD check as-is

  • compEpiTools: passes R CMD check as-is

  • SICtools: test_snpDiff() (from test_snpDiff.R) FAILS!

  • VariantTools: passes R CMD check as-is

A random sample of a few other software packages were tested

  • AllelicImbalance: migrated (changes not committed yet) passes R CMD check

  • BaalChIP: passes R CMD check as-is

  • chimera: passes R CMD check as-is

  • CoverageView: passes R CMD check as-is

  • exomeCopy: passes R CMD check as-is

  • exomePeak: passes R CMD check as-is

  • GenomicAlignments: passes R CMD check as-is

  • gmapR: does NOT compile (needs to be migrated)

  • MEDIPS: passes R CMD check as-is

  • methylPipe: passes R CMD check as-is

  • rnaSeqMap: passes R CMD check as-is

  • ssviz: passes R CMD check as-is

  • systemPipeR: passes R CMD check as-is

All workflow packages were tested

  • rnaseqGene: vignette builds as-is

  • sequencing: vignette does NOT build (because of AnnotationHub razip'ed FASTA file AH18522)

Status of CRAN packages that depend directly on Rsamtools

As of Feb 6, 2019, 8 CRAN packages depend directly on Rsamtools (i.e. via their Depends, Imports, or LinkingTo field).

  • BinQuasi: passes R CMD check as-is

  • Brundle: passes R CMD check as-is

  • ExomeDepth: passes R CMD check as-is

  • hoardeR: fails to install because of unrelated pre-existing problem:

    Error : object ‘importGFF3’ is not exported by 'namespace:GenomicTools'

    (See https://www.r-project.org/nosvn/R.check/r-devel-linux-x86_64-debian-gcc/hoardeR-00install.html)

  • NIPTeR: passes R CMD check as-is

  • PlasmaMutationDetector: passes R CMD check as-is

  • RAPIDR: passes R CMD check as-is

  • spp: passes R CMD check as-is

What to do about razip-compressed FASTA files and their index files?

The problem

RAZF is no longer supported in recent samtools, and old razip-compressed FASTA files (.rz) and their index files (rz.fai) are now causing problems. Here is an example from the sequencing workflow:

library(AnnotationHub)
ah <- AnnotationHub()
fa <- ah[["AH18522"]]
library(Rsamtools)
idx <- scanFaIndex(fa)  # still works
long <- idx[width(idx) > 82000]
getSeq(fa, param=long)  # ERROR! ('open' index failed)

What to do about it?

The new way to go is to compress with bgzip and to recreate the index.

From the Unix command line

/path/to/samtools-1.7/htslib/bgzip myfile.fa       # creates myfile.fa,gz

/path/to/samtools-1.7/samtools faidx myfile.fa,gz  # generates index files
                                                   # myfile.fa.gz.fai
                                                   # and myfile.fa.gz.gzi

Note that the compression step is actually optional i.e. one can use samtools faidx directly on the uncompressed FASTA file:

gunzip myfile.fa,gz  # uncompress to get myfile.fa back
/path/to/samtools-1.7/samtools faidx myfile.fa     # generates index file
                                                   # myfile.fa.fai only

Note that in this case, only the .fai index file is generated (no .gzi index file).

From R

library(Rsamtools)
bgzip("myfile.fa")
indexFa("myfile.fa.bgz")

Then create a FaFile object that can be used with getSeq() as usual:

fa <- FaFile("myfile.fa.bgz")

TODO

Here is an itemized summary of what still needs to happen (some details are provided above in this document):

  • Chase down old razip'ed FASTA files and update them. Places to look at:

    • AnnotationHub
    • ExperimentHub
    • data annotation packages
    • data experiment packages
    • software packages

  • Troubleshoot applyPileups() unit test failures.

  • Troubleshoot test_BcfFile_scanBcfHeader_remote() failure (test is located in inst/unitTests/test_BcfFile.R).

  • Migrate asBcf().

  • Update NEWS file.