Test on Low-Pass whole genome sequenced (LP-WGS) samples

[Sohambiswas8]

Hi, @kiranchari

I should have asked this question as well in the previous issue. My tumor samples are Low-Pass whole genome sequenced, with an average depth of coverage of 1.6x to 4x, much less than usual WGS. These are prostate cancer (PCa) individuals. Is VarNet-T tested on LP-WGS for the detection of SNVs and InDels?

I have run a single test. After the predict.py step, I have only 81 'PASS' variants, and I don't see any PCa-specific mutations like in AR, CHEK2, or other known DNA-damage repair gene mutations.

The commands are here:

1. Step 1)
"docker run -it --rm -v /home/soham:/host_data \

-w /VarNet kiranchari/varnet:latest /bin/bash \

-c "git pull; mkdir -p /host_data/soham/LP-WGS/somatic/varnet/output; python filter.py \

--sample_name LPS1 --tumor_bam /host_data/soham/LP-WGS/markdup/LPS1.markdup.bwa.bam \

--processes 4 --output_dir /host_data/soham/LP-WGS/somatic/varnet/output --reference /host_data/ref/hg38.fa"


2. Step 2) Filter (I am not sure whether this step is needed again)

"docker run -it --rm -v /home/soham:/host_data -w /VarNet kiranchari/varnet:latest \

python filter.py --sample_name LPS1 --tumor_bam /host_data/soham/LP-WGS/markdup/LPS1.markdup.bwa.bam \

--processes 4 --output_dir /host_data/soham/LP-WGS/somatic/varnet/output --reference /host_data/ref/hg38.fa"

3. Step 3) Predict

"docker run -it --rm -v /home/soham:/host_data -w /VarNet kiranchari/varnet:latest python predict.py \

--sample_name LPS1 --tumor_bam /host_data/soham/LP-WGS/markdup/LPS1.markdup.bwa.bam \

--processes 4 --output_dir /host_data/soham/LP-WGS/somatic/varnet/output --reference /host_data/ref/hg38.fa"

What can be the issue here? I haven't performed the realigning around the Indels step yet. The BAM file is only duplicate-marked.
Or, is it because of the LP-WGS samples, and hence some internal thresholds are filtering 'true variants'?

Thank you in advance,
Soham

[kiranchari]

Hi @Sohambiswas8,

That is likely happening because you are running VarNet on an lpWGS sample. VarNet was not originally designed or tested for the lpWGS setting, and it applies certain strict filters that many true mutations in your low-pass samples simply won't pass. For example, there is a hardcoded minimum coverage of 7 and a minimum number of variant reads of 2.

However, if there is a strong interest in this use case, we could potentially introduce user-defined arguments to lower these filter thresholds (specifically for coverage and minimum variant read support) to better accommodate lpWGS data. Please note that we still haven't formally tested the method's performance under these conditions. @skandlab, what are your thoughts on adding this flexibility?

In the meantime, you may also manually try modifying these filters in the code yourself, but please keep in mind that we cannot guarantee any results as we haven't formally tested the method's performance under these conditions.

Regarding your secondary question: you do not need to run the filter.py step twice.

[Sohambiswas8]

@kiranchari Thank you for your prompt answer. Can you keep this thread open? I will get back to you shortly. I am testing with LPS samples with lower thresholds.