I subset a BAM to only include primary reads using the following samtools commands:

samtools view -F 256 input.bam > input.primaryOnly.sam


Now, in order to convert this SAM into a new subsetted BAM, I would need to add a header. The way I normally do this is with Picard:

java -jar picard.jar ReplaceSamHeader \
I=input.primaryOnly.sam \
O=new.sam


whereby the header input.header.sam is from the original BAM:

samtools view -H input.bam > input.header.sam


Then I could convert the SAM into a BAM with

samtools view -S -b i new.sam > new.bam


My problem is, I'm using the original BAM header. If I look back at the BAM, it now contains wrong information, e.g.

samtools flagstat new.bam


might save that secondary alignments exist, when they don't.

How do I get a BAM header that makes sense given my new BAM (with only primary alignments)?

• "Now, in order to convert this SAM into a new subsetted BAM, I would need to add a header." I don't understand why you do ignore the sam header at the beginning ? why don't you just use 'samtools view -F 256 -b -o out.bam input.bam' ? Feb 7 '18 at 19:30
• samtools flagstat looks at the statistics for the reads, not the header. The thing that would possibly change in the header would be sequence lengths (i.e. from samtools idxstats), but as long as the reference sequence is the same for your original and new BAM, that won't be a problem.
– gringer
Feb 7 '18 at 21:30

Your first command only needs a slight modification to add in -h. This will create a SAM file with a header.

samtools view -h -F 256 input.bam > headered.input.primaryOnly.sam


-h Include the header in the output.

If your primary end goal is to create a BAM, then as Pierre has pointed out you can create BAM files directly from samtools view with the -b option:

samtools view -b -F 256 input.bam > input.primaryOnly.bam


And if your end-goal is a sorted BAM file, then pipe the BAM output through samtools sort:

samtools view -b -F 256 input.bam | samtools sort > sorted.input.primaryOnly.bam


If you have an existing BAM file, you can use Picard (as you have mentioned), or manually piece the files together with cat, which will have a similar function (although Picard probably includes a few checks to make sure that the header is consistent with the input file):
(samtools view -H input.bam; cat input.primaryOnly.sam) > headered.input.primaryOnly.sam

If the only thing that matters is that the BAM file has correct reference information (which is fairly common), then the -T option can be used with samtools view. Again, -h or -b is required to make sure both the header and the usual SAM fields are displayed:
samtools view -T reference.fa -h input.primaryOnly.sam > headered.input.primaryOnly.sam