Assertion failures in genotyping step about bayestyper

[jonassibbesen]

Hi Hannes,

Thank you for your interest in BayesTyper. I read your GraphTyper preprint recently and found it really interesting.

The cause behind the assert you are seeing is that there is not enough kmer information in order to estimate the parameters of the negative binomial distribution that is used model the kmer counts across the genome. Only kmers that are parsed and does not match a variant but matches a region in between variants are used for this estimation, which could explain why num_obs is 0 in your case.

Moreover to estimate the parameters of our noise model, the program currently uses a million SNV’s that does not overlap any other variants.

Because of these two things the optimal way to run Bayestyper is to do it on the full genome with millions of variants as input. However, if you want to test it on a smaller set I have also previously gotten good results running it on a single chromosome with only half a million variants.

I am sorry that this information was not clear from the documentation.

Please let me know if you run into any other problems.

Best wishes,

Jonas

[hannespetur]

Thanks Jonas,

I think I understand now. I will try again with the entire chromosome 20, and then the entire genome if that does not work. But the number of variants does not matter here, right? From my understanding, these two parameter estimations should work fine even if there are no real variants (i.e. the samples have two copies of the reference haplotype).

Best regards,
Hannes

[jonassibbesen]

It should not be a problem that you do not have any real variants in your input set, since BayesTyper handles reference and alternative alleles identically. However it is important that your input set contains a lot of variants, real or not, since kmers are only parsed if all smaller subsequences (18 nt) of the kmer matches one or more variants in the set. Therefore if your input set only contains a small number of variants, only a small number kmers will be parsed and used in the subsequent parameter estimations steps.

Best wishes,

Jonas

[hannespetur]

Ah, I see. Thank you.