Introduction

cram-archiver was written to help with an archival task where a substantial volume of BAM files needed to be converted to CRAM in order to save disk space.

Features:

• Automated recursive discovery of all .bam files in a directory. Symbolic links are ignored.

• Multiple reference support. Cram-archiver loads in fasta indexed references and checks that the appropriate BAM file is matched to the appropriate reference using the contig and length information in the BAM header.

• Performs CRAM conversion using samtools view.

• Performs samtools checksum --all on the BAM and CRAM file and checks if the checksum matches.

• On by default: writes checksum files for manual verification.

• On by default: writes CRAM indexes.

• Optional: deletes BAM file after conversion.

• Optional: Set a minimum age in days for the BAM file’s last modified time. If the file is “older” than the set number of days, the file will be converted.

Caveats

CRAM was never intended and built as a “pure” archival format with bit-for-bit reproducibility. As a result it is impossible to get the original BAM file back from a pure CRAM file. There are several reasons for this:

• BAM files are by definition always bgzip compressed using the DEFLATE algorithm. Differences in the DEFLATE algorithm implementation can cause different outputs.

• When converting a BAM and its derived CRAM to SAM the two SAMs can have differences too:

  • MD and NM tags are not stored in CRAM files but always calculated on the fly when decoding. If the MD and NM flags were not present in the original BAM, this can cause differences.

  • The order of tags might be different.

  • M, = and X in CIGAR strings. = means that the nucleotide is the same at this position. X means a mismatch at this position. M means that the position matches (no indels), but gives no information whether it is X or =. Since X and = can be derived from the sequence, the extra information is redundant and CRAM stores everything as M. This can give rise to differences.

  • Redundant information in BAM files such as unaligned reads with MAPQ values or CIGAR strings. This does not get stored.

  • Errors, such as wrong mate pair information. Some of it may be fixed during the CRAM conversion.

To assure the CRAM file is “functionally the same” as the BAM file, the samtools checksum tool with the --all flag is run. For more information about comparing BAM and CRAM checkout the discussion here.

Quickstart

Converting a single BAM file:

cram-archiver -r my_reference.fasta my.bam

This will create my.cram, my.cram.crai, my.cram.checksum and my.bam.checksum. Checksum file creation can be turned of with --dont-write-checksums. The checkums will still be checked, just not written to disk.

Archiving a directory with BAMs, but only BAMs that have a lost modified time older than 30 days. Also, there are hg19 and hg38 BAM files in the directory.:

cram-archiver --reference hg19.fasta --reference hg38.fasta --minimum-age-days 30 my_directory

If the --delete flag is added, all the converted BAM files will be deleted and just the CRAM files remain. This only happens when the conversion is successful and the checksums match.

Usage

usage: cram-archiver [-h] -r REFERENCE [-t THREADS] [-d MINIMUM_AGE_DAYS]

[–delete] [–cram-version CRAM_VERSION] [–exclude EXCLUDE] [–exclude-list PATH] [–dont-write-checksums] [–dont-write-index] [–dry-run] [-v] [-q] [–version] PATH

positional arguments:

PATH Path to BAM file or directory to be recursively

searched.

options:

-h, --help

show this help message and exit

-r REFERENCE, --reference REFERENCE

Reference to be used for CRAM conversion. Can be used multiple times. Reference will be checked with the BAM file.

-t THREADS, --threads THREADS

The number of threads used for conversion and checksumming.Default: 1.

-d MINIMUM_AGE_DAYS, --minimum-age-days MINIMUM_AGE_DAYS

The minimum last modification of the BAM file in days prior. This assumes the system clock timezone matches that of the file while also assuming that every day has 24x60x60 seconds. Default 0.

--delete

Delete BAM files after successful conversion.

--cram-version CRAM_VERSION

CRAM version to use for CRAM conversion. Default: 3.0.

--exclude EXCLUDE

Exclude file or directory from conversion. Can be supplied multiple times.

--exclude-list PATH

Supply a newline-separated file with files and directories to exclude.

--dont-write-checksums

Do not store samtools checksum output on disk.

--dont-write-index

Do not write index files for CRAM files.

--dry-run

Print the paths of the to be archived BAM files. Perform no actions.

-v, --verbose

Display more logging information.

-q, --quiet

Display less logging information.

--version

show program’s version number and exit

On CRAM format settings

Cram-archiver uses version 3.0 of the CRAM standard by default. The reason for this is that CRAM version 3.0 is better supported than version 3.1. CRAM version 3.1 comes with newer codecs and is able to achieve smaller file sizes because of that. For more information checkout the article on advances in CRAM by James Bonfield.

Cram archiver uses the CRAM default presets. CRAM has some presets: fast, normal, small and archive. However, the size differences between normal and archive are quite small (less than 6% smaller in our tests). On top of that, the memory requirements rise steeply especially on very long read alignments of ONT data.

Acknowledgements

A huge thank you to James Bonfield (@jkbonfield) for providing a lot of information and background about CRAM and its tooling. This was invaluable for creating this project. James Bonfield has also spent a lot of effort into making CRAM the very usable format it is today for which we are very grateful.