Concordance Between Two Identical GLIMPSE2 Runs is Less Than 1

[ethelvol21]

I observed that running the same GLIMPSE2 command twice with identical inputs and parameters does not yield perfect concordance (1.0) per sample. Despite using the same random seed, there are discrepancies in the results, as shown in the histogram below:

Histogram of Concordance (Chromosome 10):

I used the following command twice:

GLIMPSE2_phase \
    --threads {cores_n} \
    --bam-list {files_list} \
    --reference {ref_file}_{chrom}_{regs}_{rege}.bin \
    --output {WORK_PATH}/chunks/phase/{output_file}_{chrom}_{regs}_{rege}.bcf \
    -F {ref_path}Homo_sapiens_assembly19.fasta \
    --seed 42

The concordance per sample is less than 1, despite the same random seed (--seed 42) being used for both runs.
Given the same random seed and identical input data, I expected the output to be deterministic and yield a concordance of 1.0 per sample.

Are there any specific settings or flags I should use to ensure deterministic output, or is there an inherent non-deterministic component in GLIMPSE2?

[srubinacci]

Hi,

Using multi-threading makes the output non deterministic, due to random invocation of the RNG. Running the same command on a single thread should solve your issue.