issue with automatic sex detection in cnvkit.py batch: female samples labelled as male

[eesiribloom]

I have 20 female samples. I run this command

cnvkit.py batch ${BAM_INPUT} \
--seq-method wgs \
--drop-low-coverage \
--reference ${REFERENCE_CNN} \
--scatter --diagram \
-d ${OUTPUT_DIR} \
-p ${THREADS} \
--diploid-parx-genome grch38

and I get these outputs from automatic sex detection but I cant see why some samples are detected as male based on these values, they all seem quite similar to me

Relative log2 coverage of chrX=-0.14, chrY=-21.4 (maleness=0.0769 x 0.962 = 0.074) --> assuming female
Relative log2 coverage of chrX=0.224, chrY=-20.4 (maleness=0.114 x 0.951 = 0.108) --> assuming female
Relative log2 coverage of chrX=-0.132, chrY=-20.9 (maleness=0.135 x 0.938 = 0.127) --> assuming female
Relative log2 coverage of chrX=-1.05, chrY=-21.3 (maleness=93.2 x 0.952 = 88.7) --> assuming male
Relative log2 coverage of chrX=-0.324, chrY=-22.2 (maleness=0.464 x 0.963 = 0.447) --> assuming female
Relative log2 coverage of chrX=0.237, chrY=-21.4 (maleness=0.111 x 0.954 = 0.106) --> assuming female
Relative log2 coverage of chrX=-0.843, chrY=-23.1 (maleness=25.1 x 0.957 = 24) --> assuming male
Relative log2 coverage of chrX=-0.349, chrY=-21.7 (maleness=0.492 x 0.946 = 0.466) --> assuming female
Relative log2 coverage of chrX=-0.838, chrY=-22 (maleness=16.2 x 0.957 = 15.5) --> assuming male
Relative log2 coverage of chrX=-0.501, chrY=-22 (maleness=1.2 x 0.94 = 1.13) --> assuming male
Relative log2 coverage of chrX=-1.21, chrY=-23.3 (maleness=8.33 x 0.952 = 7.93) --> assuming male
Relative log2 coverage of chrX=0.115, chrY=-22.2 (maleness=0.0258 x 0.957 = 0.0247) --> assuming female
Relative log2 coverage of chrX=-0.104, chrY=-23 (maleness=0.0343 x 0.954 = 0.0328) --> assuming female
Relative log2 coverage of chrX=-0.183, chrY=-22.3 (maleness=0.133 x 0.943 = 0.126) --> assuming female
Relative log2 coverage of chrX=-0.333, chrY=-23 (maleness=0.396 x 0.946 = 0.375) --> assuming female
Relative log2 coverage of chrX=-0.0119, chrY=-20.2 (maleness=0.00489 x 0.937 = 0.00458) --> assuming female
Relative log2 coverage of chrX=-0.055, chrY=-23.3 (maleness=0.00616 x 0.952 = 0.00587) --> assuming female
Relative log2 coverage of chrX=-0.496, chrY=-20.9 (maleness=0.984 x 0.94 = 0.925) --> assuming female
Relative log2 coverage of chrX=0.117, chrY=-20.8 (maleness=0.0218 x 0.946 = 0.0206) --> assuming female
Relative log2 coverage of chrX=-0.891, chrY=-22.7 (maleness=118 x 0.954 = 112) --> assuming male

[etal]

It looks like the chrX coverages are low in a few of these samples; if it drops below about -0.5 or so then the crude method CNVkit uses for detecting genetic sex will call it male. The method is probably being too sensitive to low X in the cases where chrY is also very low (which is true for all these samples); I'll take a look at retuning it, and also exposing user-provided sample sexes as an input.

[eesiribloom]

Will it strongly affect the end results if some of the samples are incorrectly assigned as male given it's a pooled reference?

If so and I want to ensure all the samples are classed as female which steps do I need to run outside of the batch command?

Is it just cnvkit.py reference where the sex is automatically detected?