Why does running STITCH twice with the same parameters result in differences in SNP genotyping for many SNPs?

[thanhtruc66]

Hi Author,
I ran STITCH twice with the same parameters (K = 8, nGen = 100) and got two VCF files with the same number of SNPs on chromosome 1 of Sus scrofa. However, many of the SNPs have different genotyping results: 85,774 out of 1,558,794 SNPs (5.5%). Could you please explain why this might happen?

[saulpierotti]

Hi,
thank you for your question and sorry for the late reply. STITCH is a separate tool from my pipelines so this issue would be best addressed by the STITCH authors at https://github.com/rwdavies/STITCH.

Anyways, this is expected because STITCH randomly initialises the haplotypes before the EM process, so you can get slightly different results each run. This can be avoided by setting a random seed in R. However, if the result is drastically different each time this reflects the fact that STITCH is not converging properly, and parameter adjustment/more samples/higher sequencing depth may be needed.