gatk_cnv step 3 has the potential of attempting to interpret log2 of 0. This is a rare edge case where the count for the interval is 0 and is not filtered out in any of the previous steps.
|
gatk --java-options "-Xmx12g" CreateReadCountPanelOfNormals \ |
|
--tmp-dir {{ temp_dir }}/temp3/ \ |
|
--input {{ temp_dir }}/{{ pair.normal.name }}.{{ aligner }}.counts.hdf5 \ |
|
--minimum-interval-median-percentile 0.0 \ |
|
--annotated-intervals {{ temp_dir }}/anno_preprocessed.interval_list \ |
|
--output {{ temp_dir }}/cnv.pon.hdf5 |
|
|
|
{# Make working and temp directories #} |
|
mkdir -p "{{ temp_dir }}/temp4" |
|
|
|
{# Denoise the tumor using the panel of normals made from the matched normal #} |
|
gatk --java-options "-Xmx12g" DenoiseReadCounts \ |
|
--tmp-dir {{ temp_dir }}/temp4/ \ |
|
--input {{ temp_dir }}/{{ pair.tumor.name }}.{{ aligner }}.counts.hdf5 \ |
|
--count-panel-of-normals {{ temp_dir }}/cnv.pon.hdf5 \ |
|
--standardized-copy-ratios {{ results_dir }}/{{ pair.name }}.{{ aligner }}.standardizedCR.tsv \ |
|
--denoised-copy-ratios {{ results_dir }}/{{ pair.name }}.{{ aligner }}.denoisedCR.tsv |
One solution to this is setting the --minimum-interval-median-percentile to be really small, since this can use scientific notation a value such as 1E-30 would work and avoid 0's being passed into a log2 calculation.
