Nondeterminism in PULSAR_1.0.pl across different runs / machines

[lilybhattacharjee5]

Hi,

First of all, thank you for writing these scripts! I'm a grad student at UC Davis who is currently using them for a research project. I noticed that for PULSAR_1.0.pl, running with the same parameters on the same input files (e.g. the given ones in Example/) lead to different Phased_chromosome.vcf outputs. Between runs, I deleted the old Phased_chromosome.vcf / Potential_private_variants.rv outputs.

Run 1:
command used: perl PULSAR_1.0.pl Example/Seven_sib_pedigree.csv Example/Genotypes.vcf 5 Example/Allele_frequency.csv 0.05

Run 2:
command used: same as Run 1

The diff for the generated VCF files is not empty:

Comparing the two files in R gives different numbers of NaNs (& as shown in the diff, some column values are different as well):

all.equal(phased1@gt, phased2@gt)
"'is.NA' value mismatch: 45 in current 147 in target"

print(sum(is.na(phased1@gt)))
147
print(sum(is.na(phased2@gt)))
45

all.equal(phased1@fix, phased2@fix)
TRUE
all.equal(phased1@meta, phased2@meta)
TRUE

Similarly, I noticed that scp-ing the Example folder / given PULSAR_1.0.pl script to a remote machine gave even more different results:

Remote run:
command used: same as runs 1/2

all.equal(localPhased@gt, remotePhased@gt)
"'is.NA' value mismatch: 21 in current 147 in target"

print(sum(is.na(localPhased@gt)))
147
print(sum(is.na(remotePhased@gt)))
21

all.equal(localPhased@fix, remotePhased@fix)
TRUE
all.equal(localPhased@meta, remotePhased@meta)
"1 string mismatch"

As I'm not very familiar with Perl, I was wondering if this is expected behavior? Should there be some degree of nondeterminism in outputs across runs or do you have an idea of what may be causing this?