Missing SNPs from high depth file

[danielwood1992]

Hello,

I am using a relatively high depth (~200X) ONT bam file from an individual to call SNPs. When I use the full depth file, I only get ~182,000 SNPs in the merged.vcf.gz file - if I downsample the bam file to ~100X, I get ~14.7M SNPs merged.vcf.gz file. My understanding is that Clair3 should downsample excessively high depth regions, so I'm not sure why this behaviour is happening - is there a parameter I should be changing, or a maximum depth I should downsample the bam file to?

I am using Clair3 v1.0.11 with the following command:

apptainer exec -B $indir,$outdir $sif /opt/bin/run_clair3.sh --include_all_ctgs --bam_fn=$bam_name --ref_fn=$genome --threads=${NSLOTS} --platform="ont" --model_path="/opt/models/r941_prom_sup_g5014" --output=$outdir --gvcf --var_pct_full=1 --ref_pct_full=1

Any advice on this would be much appreciated!

[zhengzhenxian]

Hi,

This outcome is unexpected. Our full-alignment model has a maximum depth of 89, so we would expect similar subsampling results for your 200x and 100x datasets. Could you please send your run_clair3.log file and the log folder from your output directory to my email zxzheng@cs.hku.hk for us to pinpoint the issue? Thanks~

Zhenxian

[danielwood1992]

Hi Zhenxian, sorry for the delayed response on this - I've sent an email to you with the log files. Thanks!

[danrdanny]

I may have a similar issue. I'm happy to open a new thread if recommended. Briefly: we're benchmarking NA12878 and miss a few known SNVs when we have high-depth data (>150x) that we don't miss with lower coverage data (~30x). Using Clair3 1.0.8. Command is:

/opt/anaconda3/envs/clair3-1.0.8-nogpu/bin/run_clair3.sh --ctg_name=$chrList --bam_fn=$bamFile --ref_fn=/data/dat/hg38/hg38.no_alt.fa --threads=50 --platform=ont --model_path=r1041_e82_400bps_sup_v500 --output=temp --enable_phasing

Example of missed SNV is shown in IGV screenshot which shows squished phased reads. The het C>G is correctly called in both (high coverage on top, low coverage on bottom). But, the het T>C is not called in the high coverage sample on top while it is called on the bottom.