Hi,

I'm developing a Nextflow workflow that integrates multiple CNV callers for improved CNV calling from exome sequencing data.

The tools I have been working with so far are: CANOES, CLAMMS, XHMM, CNVkit, INDELible, GATK-gCNV and DRAGEN Germline Enrichment.

I had read how combining the outputs of multiple CNV callers improves sensitivity and precision, and so I have decided to merge the outputs of these different tools with Truvari. To this end, the outputs of CANOES, CLAMMS and INDELible had to be converted to VCF format.

The problem with Truvari is that it retains the VCF entry of the CNV call with the highest value in the QUAL column. The problem with this is that all the callers have different statistical and probabilistic algorithms as to how they call CNVs and the scales of the quality scores they produce are not comparable.

As such, the filtering guidelines for each tool differ greatly. For example, it is recommended to filter the CANOES output to retain calls with Q_SOME > 80, while for CLAMMS, it is recommended to retain calls with Q_SOME > 500 & Q_EXACT > 0.

I asked Claude for advice and it suggested normalising the scores and adding the normalised quality score to the QUAL column. I'm unsure as to how to do this exactly?

If anyone has advice or ideas or guidance, it would be much appreciated.

Kind regards,

Yoni Wolberg