Strange bug(?) in GATK

[alexbmp]

Just to see if anybody knows...

I got this funny error(?).

I ran GATK for chr1 (with -L chr1_startend.bed) for all of my reads in bam file,

then I ran GATK for hg18.fa (with no -L option) for all of my reads in bam file (the same sample as when I ran GATK with only chr1;
by the way, -L option sets the interval you want to traverse-walk with GATK,
it's known to be completely independent between chromosomal processes).

For chr1-only-run, I got 40,179 SNPs/indels,
while for all-chr-run, I got 40,167 SNPs/indels.

I've chased the read counts of chr1 in both bam file processes;

When I start using GATK the number of reads differ between chr1-only and all-chr's processes.
(e.g. I have 9,468,691 chr1-only-file chr1 reads after realignment,
and 9,468,707 for all-chr-file chr1 reads; a small difference, but A difference)

According to GATK documentations, this doesn't make any sense.

SG works very efficiently in the GATK, provided the output of a walker is independent per site (such as the Unified Genotyper) or per chromosome (Table Recalibrator or Indel Realigner).

(http://www.broadinstitute.org/gsa/wi...m_and_the_GATK)

Does anybody have any ideas?

Anyways, have a nice day all!