Comparison with samtools

Functionality

Viewing SAM/BAM

Feature	sambamba view	samtools view	Notes
BAM support	Full	Full
SAM support	Full	Full	sambamba skips (syntactically) invalid tags and sets invalid fields to default values
Error messages	Descriptive	Incomplete	Where samtools says just 'truncated file', sambamba prints detailed error message with a description what is wrong with BAM file
Multithreaded BAM decompression	Yes	No
Non-seekable file stream support	Yes	Yes
Skipping invalid reads	Optional	No	Sambamba library also includes a module for creating custom validation tools
Filtering	Powerful	Limited	sambamba view comes with a simple query language for filtering alignments
JSON output	Yes	No	useful for interacting with scripting languages
Progressbar	Optional	No

Other tools

Feature	sambamba	samtools
Indexing	Yes, multithreaded	Yes, single-threaded
Merging BAM files	Yes, multithreaded decompression and compression	Yes, compression is multithreaded
Automatic SAM header merging	Yes	No
Multithreaded BAM file external sort	Yes	Yes
Flag statistics	Yes, multithreaded	Yes, single-threaded
(other utilities available in samtools are not implemented in sambamba)

Performance

Here are some benchmarks on two configurations:

• Intel Atom N450 @ 1.66GHz (1 core with hyperthreading), 1GB of RAM
• 2x Intel Xeon E5310 @ 1.60GHz (8 cores without hyperthreading), 8GB of RAM

On both machines, sambamba was built with GDC compiler (which is used for building debian packages), and samtools was built with its default makefile using gcc -02.

Tools were tested on HG00125.chrom20.ILLUMINA.bwa.GBR.low_coverage.20111114.bam (denoted by $FILENAME in command lines), 301MB in size.

Indexing BAM file (empty file cache)
sambamba index $FILENAME				samtools index $FILENAME
Configuration	Time	Memory usage	CPU load	Configuration	Time	Memory usage	CPU load
Intel Atom N450	12.29s	32MB	147%	Intel Atom N450	13.6s	1.4MB	92%
2x Intel Xeon E5310	6.96s	32MB	139%	2x Intel Xeon E5310	8.73s	1.4MB	93%
Indexing BAM file (file fully cached into RAM)
sambamba index $FILENAME				samtools index $FILENAME
Configuration	Time	Memory usage	CPU load	Configuration	Time	Memory usage	CPU load
Intel Atom N450	9.43s	32MB	188%	Intel Atom N450	12.08s	1.4MB	99%
2x Intel Xeon E5310	2.21s	32MB	433%	2x Intel Xeon E5310	7.98s	1.4MB	100%
Filtering reads from a region, with BAM output (empty file cache)
sambamba view -f bam $FILENAME 20:10,000,000-20,000,000 -F "mapping_quality >= 50" -o test.bam				samtools view -b $FILENAME 20:10,000,000-20,000,000 -q50 -o test.bam
Configuration	Time	Memory usage	CPU load	Configuration	Time	Memory usage	CPU load
Intel Atom N450	22.96s	90MB	98%	Intel Atom N450	23.16s	1.8MB	96%
2x Intel Xeon E5310	5.24s	90MB	250%	2x Intel Xeon E5310	10.83s	1.8MB	98%
Counting reads from a region (file fully cached into RAM)
sambamba view $FILENAME -c -F "[RG] == 'ERR016156' and proper_pair and first_of_pair and not duplicate" 20:1000000-3000000				samtools view $FILENAME -c -r 'ERR016156' -f66 -F1024 20:1000000-3000000
Configuration	Time	Memory usage	CPU load	Configuration	Time	Memory usage	CPU load
Intel Atom N450	0.53s	50MB	144%	Intel Atom N450	0.42s	1.3MB	99%
2x Intel Xeon E5310	0.20s	50MB	208%	2x Intel Xeon E5310	0.27s	1.5MB	100%

Conclusion

As you can see, sambamba exploits parallelism where samtools does not. The faster the storage you use, the more the speedup is (see results for indexing).

However, there're some drawbacks at the moment. Memory usage is higher due to extensive use of various buffers, and region queries are slower in some cases (though not much, about 10-20%).