Unexpected high MLreject rate with Twist Exome

[giovannabloise]

Hi,

I processed Twist Exome samples with the DNAScope pipeline using the ML model. I see that about 15% of variants are marked as MLreject across multiple samples.

Is this expected? And is the ML model suitable for Twist Exome data, or should I disable it?

Thanks!

Sample	MLrejected	Total	%
Sample1.dnamodelapply.vcf.gz	16223	107336	0.151142208
Sample2.dnamodelapply.vcf.gz	16390	106117	0.154452161
Sample3.dnamodelapply.vcf.gz	15968	108462	0.147222069
Sample4.dnamodelapply.vcf.gz	15791	102785	0.153631366
Sample5.dnamodelapply.vcf.gz	15433	105988	0.145610824
Sample6.dnamodelapply.vcf.gz	16027	104365	0.153566809

[DonFreed]

This can be expected. With a model file, DNAscope will run in a high sensitivity mode to call additional low-quality sites as "variant candidates". These low quality sites will then be filtered in the subsequent DNAModelApply step.

The ML model (in the Illumina WES model bundle) is suitable for Twist Exome data.

Be sure to restrict variant calling to the capture targets using the --interval and --interval_padding arguments during the DNAscope step to avoid calls from off-target sites.