[HESmith]

Using 'samtools index' and then 'samtools idxstats' on a sorted file will give you total counts for mapped and unmapped reads, which is slightly easier to check compared to looking at a few million lines to find what's missing.

Thanks, David. I hadn't thought of that.

[KevinLam]

What HESmith said, but I'll add that if you really want to be 110% sure that nothing was lost/changed, you can use bamHash on both files. If the checksums are the same then they contain the same reads (just in a different order).

https://github.com/DecodeGenetics/BamHash

is this the bamhash you are referring to?
I wonder how fast it is ...

http://www.html5rocks.com/en/tutoria...t-compression/ there's a hint in there why the zlib compression makes a smaller file on sorted files.

[dpryan]

Yup, that's the tool. It's about as fast as one would hope for a tool that has to process every alignment. Note that this isn't something you would run all the time, the original purpose was to make absolutely certain that BAM files contain all of the information in the fastq files that made them (there are functions to calculate equivalent hashes of fastq files) so the fastq files could be deleted.

But anyway, you're correct that sorted files are easier to compress, since individual blocks will contain more repetitive information.

[gringer]

If you're interested in conserving space, there's also a compressed BAM format that only records differences from the reference alignment. Running 'samtools view -C -T <reference.fasta> <input.sam|input.bam>' will produce such a file, assuming you're using a sufficiently new release of samtools.