bug in samtools view -s?

[jkzebrafish]

Hi, I'm experiencing difficulties trying to downsample .bam files using samtools view -s. Specifically some of the commands fail while others work; this seems sometimes to be correlated with the -s float argument being > 0.5 (but not always). Here I'm c/p'ing some of the code that worked and some which failed.

Thanks to any helpful suggestions!

samtools view -b -s 0.271 1.bam > 1_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.5077 2.bam > 2_ds.bam # gives 0 reads unexpectedly
samtools view -b -s 0.2113 3.bam > 3_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.3322 4.bam > 4_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.5306 5.bam > 5_ds.bam# gives 0 reads unexpectedly
samtools view -b -s 0.204 6.bam > 6_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.3841 7.bam > 7_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.4691 8.bam > 8_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.6861 9.bam > 9_ds.bam # gives 0 reads unexpectedly
samtools view -b -s 0.2261 10.bam > 10_ds.bam # gives 730697 reads as expected

samtools view -b -s 0.6653 23.bam > 23_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.0444 24.bam > 24_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.0492 25.bam > 25_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.1648 26.bam > 26_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.0801 27.bam > 27_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.171 28.bam > 28_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.0979 29.bam > 29_ds.bam # gives 730697 reads as expected
samtools view -b -s 0.0511 30.bam > 30_ds.bam # gives 730697 reads as expected

[GenoMax]

Not answering your question directly but you could use "reformat.sh" from BBMap suite to do this as well. You can specify sampling parameters with more granularity (even as certain number of reads etc).

[cstack]

Hi, I'm experiencing difficulties trying to downsample .bam files using samtools view -s. Specifically some of the commands fail while others work; this seems sometimes to be correlated with the -s float argument being > 0.5 (but not always). Here I'm c/p'ing some of the code that worked and some which failed.

It seems there are specific read alignments that are causing the failures. You could confirm this by using taking one of the .bam's that failed, use different random seeds w/ a small sample fraction, and you should see the failure some percentage on of the time.

Are these alignments of very long reads? (> 65k bp). Alignments with cigar strings longer than the 16-bit integer limit (65,535) can behave strangely

[jkzebrafish]

Thanks cstack for the response. These are paired end 75bp reads, nothing crazy.

Here is a little more information:

samtools view -b -s 0.6861 9.bam > 9_ds.bam # gives 0 reads
samtools view -b -s 0.4861 9.bam > 9_ds.bam # gives ~50k reads
samtools view -b -s 0.5 9.bam > 9_ds.bam # gives ~50k reads
samtools view -b -s 0.5001 9.bam > 9_ds.bam # gives 0 reads
samtools view -b -s 1.6861 9.bam > 9_ds.bam # gives 0 reads
samtools view -b -s 5.6861 9.bam > 9_ds.bam # gives 0 reads
samtools view -b -s 100.6861 9.bam > 9_ds.bam # gives 0 reads

No errors or warnings are given, hence my confusion. Thanks for any insight.