I'm currently creating a pooled normal samples using 15 blood normal .bams from TCGA. The reference file generated below is used to generate the final .cns files after accounting for low coverage regions.

The issue here is that my output looks very noisy and I wonder if unmatching target .bed file is responsible for such behavior. Would it be a better approach to use flat reference or pooled tumor samples instead?

Also, should I be concerned of the following message (or similar)? :

Targets are 6.9 x more variable than antitargets

cnvkit.py coverage [normal_1.bam,...,normal_15.bam] target.bed -o [normal_1_t.cnn,...,normal_15_t.cnn]

cnvkit.py coverage [normal_1.bam,...,normal_15.bam] antitarget.bed -o [normal_1_at.cnn,...,normal_15_at.cnn]

### Pool target & antitarget .cnn files
cnvkit.py reference -x f -f hg38.fa -o tcga_normal_15.cnn *.cnn