make_truth_vcf misses truth SNPs if they lie near dropped amplicons

[Danderson123]

Hey Martin!

I ntoiced a weird occurrence when running covid-truth-eval on artic assemblies built from simulated read sets, whereby a true SNP in the VCF of the artic assembly is incorrectly being identified as a false positive when it is actually present in the simulated assembly and therefore should be in the truth VCF. Turns out the SNP was in close proximity to a run of Ns in my simulated assemblies (to mask dropped amplicons) and was being missed by varifier make_truth_vcf. I have uploaded an example truth and artic assembly with the SNP, a truth VCF with the SNP missing and the output of covid-truth-eval.

Thanks and let me know if you need any more info,
varifier_example.tar.gz

Daniel

[martinghunt]

Can you please list the commands I should run to reproduce this from your data, and what the SNP is that's causing the problems?

[Danderson123]

Hey Martin,

The command I am running is singularity run singularity/varifier/varifier.img make_truth_vcf --global_align --global_align_min_coord 25 --global_align_max_coord 29827 masked_truth_assembly.fasta reference_genome.fasta truth_vcf/1_a and the SNP that is missed is 3037.

[martinghunt]

Are you using MN908947.3 for reference_genome.fasta?

Position 3037 in MN908947.3 aligns inside a gap in masked_truth_assembly.fasta. Screenshot attached - 3037 is the C highlighted in the middle.

.