Lofreq viterbi

[rahil19]

Hi,
I recently used Lofreq v 2.1.5 to make variants calls to the HIV WGS data. Before running lofreq I aligned reads to BWA and filtered for properly paired alignment using samtools. After running lofreq commands in the following series:
viterbi --> indelqual --> alnqual --> call, I noticed made some frameshift mutation calls. Upon looking at the alignment on IGV at one of the frameshift mutation region, the top one shows before the lofreq preprocessing and the bottom one after lofreq preprocessing. As you can see the viterbi step introduces insertion and deletion on the same reads resulting in 2 frameshift insertions and deletions reported on 29% of the reads as shown below:

Sample	HGVS.g	HGVS.c	HGVS.p	lofreq	Variant_Type	lofreq_Var_Count
A	NC_001802.1:g.5212_5213insCC	HIV1gp4:c.108_109insCC	vpr:p.Ile37fs	0.290914	frameshift_variant	3269
A	NC_001802.1:g.5214_5215delTT	HIV1gp4:c.111_112delTT	vpr:p.Ile37fs	0.295586	frameshift_variant	3268

Because insertion and deletion are present on the same read it looks more like an artifact than real. How do I fix this? Should I be removing the viterbi step? If so, do I still keep the indelqual and alnqual steps?

I've also found regions where the alignments were completely missing due to BWA (2nd figure attached) and I was wondering if I provide lofreq with raw alignment bam file (containing unmapped reads) instead of filtered proper paired alignment bam, viterbi step can possibly realign the unmapped reads in those regions with gaps?

[ammaraziz]

The docs say viterbi is not recommended for illumia data.

Did you perform the gatk best practices step before running lofreq?

[ebioman]

Sorry but is it though ?
The docs says:

Probabilistic realignment of your already mapped reads, which corrects mapping errors (run right after mapping). Not recommended for non-Illumina data.

For me that's the opposite: only use it if you have Illumina data

[ammaraziz]

Apologies you are correct. I think I misread the docs or that's a typo thinking jt was nanopore data.