coverage uniformity with GCI

[fangdongming]

Hi Quanyu, thanks for the great tool!
we assess the coverage uniformity of different sequencing data, include ONT and HiFi. we change the -mq parameter and the result deviate from our expectations. We assumed that it would cover more regions with -mq 0 than with higher thresholds , but the trend we saw changed a lot. How should we interpret this?

Thank you.

Our analysis script only modifies -mq and -d between runs:

Below are the results obtained under different -mq settings:

[yeeus]

It's so weird! Could you give me the logs or the BAM/PAF files, so that I can perform some debugging.

[yeeus]

Sorry for the delayed reply and thanks for the detailed report and the plots — what you observed can be expected given how GCI uses -mq.

In GCI, -mq/--map-qual is not just a “keep more reads → cover more bases” switch. Lowering -mq (e.g., to 0) brings in many low-MAPQ alignments, which are typically multi-mapping reads from repetitive/low-complexity regions. When the candidate alignments increase, GCI’s logic for selecting a read’s “primary/representative mapping” (especially when parsing PAF and assigning a primary target based on a score) can change substantially. As a result, some reads that previously supported one region/chromosome may be reassigned to an alternative repetitive locus (i.e., the mapping is “stolen” by another target), which makes the overall coverage distribution less uniform and can produce non-monotonic trends when you vary -mq.

So, -mq 0 does not necessarily mean “more uniquely informative coverage”; it often means “more ambiguous mappings,” which can distort uniformity metrics.

Therefore, please treat -mq as a knob controlling mapping ambiguity, not simply sensitivity. For uniformity analysis, using a strict MQ threshold (e.g., 30-60) is often more interpretable.