Proposed metrics changes to support discontiguous long reads technologies

[d-cameron]

The proposed changes leave the following still unresolved:

• yield_bp_q30 and yield_bp_read_q10_q20_q30 redundantly define the BQ30 yield
  • Proposed resolution: refactor yield_bp_read_q10_q20_q30 into yield_bp_q10 and yield_bp_q20
• read_length_percentiles_p25_p50_p75 and median_read_length both define the median read length
• Proposed resolutions: Either
  • Define: median_read_length, read_length_p25, read_length_p75
  • Define: read_length_percentiles_p25, read_length_percentiles_p50, read_length_percentiles_p75
• small_indel_count is very similar to count_deletions and count_insertions.
  • Proposed resolution: no change - keep all three.

[justinjj24]

yield_bp_q30 vs. yield_bp_read_q10_q20_q30:
These are distinct metrics, not redundant. yield_bp_q30 measures per-base quality BQ 30 or greater, whereas yield_bp_read_q10_q20_q30 measures cumulative yield from reads meeting a read-level mean quality threshold (e.g., ONT qs tag / PacBio rq tag). Not refactor this into base-level metrics?

Median read length vs. P50:
Agreed they are identical. Happy to adopt the proposed split (median_read_length, read_length_p25, read_length_p75) or standardize under read_length_percentiles_*.

Indels count:
Agree that small_indel_count is not part of the proposed metrics, so we can drop it and retain count_insertions and count_deletions.

[balaONT]

yield_bp_q30 vs. yield_bp_read_q10_q20_q30

I agree with Justin that these are distinct metrics.

yield_bp_q30

This metric represents the cumulative number of bases with a base quality score ≥ Q30. It is typically used for short-read sequencing, where base-level quality filtering is an established and commonly used practice.

yield_bp_read_q10_q20_q30

This metric is calculated at the read level. It represents the cumulative number of bases from all reads whose mean read Q score meets a specified threshold (e.g., Q10, Q20 or Q30).

Importantly:

The Q score threshold is applied to the mean quality score of each read, rather than to individual bases.
Once a read meets the threshold, all bases from that read are included in the yield calculation.
Therefore, a read that passes the threshold can still contain individual bases with Q scores below that threshold.
Read-level Q-score filtering is an established and commonly used approach for contiguous long-read sequencing, and therefore represents a distinct QC metric from base-level Q30 yield.

In this sense, yield_bp_read_q10_q20_q30 serves a similar QC purpose to yield_bp_q30, but the quality threshold is applied at the read level rather than the base level.

So, I would retain both metrics: they describe high-quality yield using two different quality-filtering approaches that reflect established practices for their respective sequencing technologies.