[etal]

Two gene-name consumers keyed on a bin's whole gene field instead of the individual names it comma-joins, so a bin covering two genes became a third gene that does not exist, and both still used the first-to-last-occurrence span rule that #1173 and #1177 removed from by_gene and scatter --gene. cnvlib/reports.py::get_gene_intervals feeds the released breaks table; cnvlib/plots.py::gene_coords_by_range feeds scatter -c chr:start-end labels. Both now group bins into loci with cnvlib.cnary.gene_runs, so one rule serves genemetrics, the diagram, scatter --gene, scatter -c region and breaks.

Both functions date to the 2014 import, when target's shortest_name collapsed multi-accession labels to one name and CNVkit never emitted a comma-joined gene value. Whole-field keying was correct by construction and became wrong in 203ed6d (2016-12-10), when annotation started flowing through join_strings. Nothing defends it: the maintainer's answer in #15 (2015) is that a FOO,BAR bin "really represent[s] 2 or more genes", and #107 → aef8b78 (2016-05-13) chose one row per real gene, with overlapping spans, for the sibling gainloss report. The breaks path was simply never revisited; #220 (2017) is a user asking what a Y_RNA row means, answered in prose rather than in code.

Decisions

A co-binned breakpoint is reported once per gene, not once under a joined name. That is #107's rule, and the gene column's documented use is breaks Sample.cnr Sample.cns | cut -f1 | sort -u > gene-breaks.txt, whose stated purpose is a list of gene names and whose sort -u absorbs the duplication. Under the old rule a reader filtering that file for CDKN2A found nothing, because CDKN2A's breakpoints were filed under CDKN2A,RP11-145E5.5.

Plot labels do the opposite where spans coincide. Genes whose runs have identical bounds share one stripe under a joined label, because two identical axvspan stripes with two text calls at the same coordinates overprint. This is the mechanics gene_coords_by_name adopted in #1177 for #458, and it is the only point where the two surfaces diverge: a table row is filtered, a stripe is looked at.

probes_left and probes_right count every bin in the locus's run, including off-target bins and backbone baits it absorbs. A copy number alteration affects a genomic region regardless of the gene model, so a CGH backbone bin or an antitarget bin inside a gene supports a breakpoint there as well as a targeted bin does. genemetrics already counts exactly those bins in its probes column, and the two sides now sum to it; before, 19 of 58 amplicon rows disagreed with it, and 5 more disagreed because their gene did not exist. test/formats/regression/p2-9_2.cnr is the UCSF500-shaped case: 1120 CGH bins and 12451 Background, 37 of them inside a real gene's run.

A trailing comma in the gene field names nothing. 171 bins of test/formats/wgs-chr17.cnr are labeled MIR3186,MIR4740,, which splits to a third, empty name; if gene: skips it at both sites, mirroring reports.group_by_genes. Adding "" to the tuple handed to gene_runs was rejected: it would make such a bin absorbable rather than run-terminating, diverging from by_gene — a subtler disagreement than the one being fixed.

Measured

Baseline worktree at 8e025ea, both trees discriminated by cnvlib.__file__ under the exact invocation.

fixture	rows before	rows after	comma-named rows
breaks amplicon.cnr amplicon.cns	49	58	9 → 0
the same at -m 4	30	38	6 → 0
breaks regression/p2-9_2.cnr (comma-free)	42	42	0 → 0
breaks wgs-chr17.cnr, 83 block-mean segments	93	218	48 → 0

On amplicon all five invented names disappear (APC,CTC-554D6.1, ATM,C11orf65, GOPC,ROS1, CDKN2A,RP11-145E5.5, KDR,RP11-530I17.1), the documented cut -f1 | sort -u list goes from 26 names to 29 real ones, and real genes recover the bins the phantoms held: APC chr5:112090792 (2, 10) → (2, 39), ATM chr11:108143413 (22, 18) → (22, 43), SMAD4 chr18:48592058 (2, 4) → (9, 4). On p2-9_2 the same 42 breakpoints are reported and two rows' counts change, PTPRD chr9:8733906 (35, 2) → (35, 10) and chr9:9003938 (36, 1) → (37, 8), both from absorbed CGH bins — that fixture isolates the counting decision from the comma fix. On wgs-chr17, Y_RNA had one 39.1 Mb interval (chr17:7,440,413-46,556,560) built from the 3 bins whose whole field was exactly Y_RNA, though 7706 bins name it; every segment boundary inside that span was a phantom row, 38 of them, and they sorted high because the ranking key is min(probes_left, probes_right). They become 6 real rows, and the widest interval on the chromosome falls to 1.18 Mb.

For labels, amplicon chr17 goes from [ERBB2, 'ERBB2,MIR4728'] to [ERBB2, MIR4728], and chr5 from two APC boxes split by a gap to one contiguous locus 112,043,331-112,179,950 beside CTC-554D6.1's own span; renderings were compared in both trees. Over all of wgs-chr17 the widest region falls from 39.1 Mb to 1.18 Mb, and -c chr17 now agrees with -c chr17 -g Y_RNA on all 13 loci in the window they share.

get_gene_intervals also got faster despite producing 76% more entries on wgs-chr17 (0.203 s → 0.144 s), the per-row itertuples loop being gone, and vectorizing the two probe-side counts takes get_breakpoints' scan from 0.055 s to 0.024 s there.

Review findings fixed

• probes_left/probes_right kept a list-era idiom over what is now an ndarray view, and computed the second count as the complement of the first the long way; verified byte-identical on three fixtures across min_probes in {1, 2, 3, 5, 10}.
• The label join was written twice; it is now one _join_labels helper, which states the overprint rule once.
• The probes equality claimed in doc/reports.rst is true of genemetrics without -s and false with it, since -s splits a gene at each breakpoint. Qualified.
• The locus rule now precedes the column list it is a premise for.
• The new cross-check against genemetrics inherited min_probes=3, dropping 25 of 79 amplicon loci from its lookup table and reaching a KeyError rather than an assertion for any gene below it.
• ignore += params.ANTITARGET_ALIASES grew a caller-supplied list in place, by two entries per call, at both sites. The same shape survives at cnvlib/segmentation/__init__.py:464, on the .cns gene-column path, and is left for its own change.

Tests

Nine new cases. Four in test/test_analysis.py pin the gene column against commas, one row per co-binned gene at one breakpoint, probes_left + probes_right against genemetrics, silence for a boundary in the gap between two loci of one name, and a placeholder bin inside a gene counting toward its support. Five in test/test_plots.py pin per-gene labeling, joint labeling for a shared locus, agreement with gene_coords_by_name over the same window, placeholder absorption, and the trailing-comma name. All four breaks cases and three of the five label cases fail against 8e025ea; the other two pin decisions rather than regressions, and were checked against wrong implementations instead — naming-bin-only counts fail on ATM (40 ≠ 65), unjoined per-gene regions fail the shared-locus case.

Full suite green, including the slow marker's frozen end-to-end pipeline; mypy and the doc build clean.

[codecov]

Codecov Report

❌ Patch coverage is 95.45455% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 76.39%. Comparing base (8e025ea) to head (b7fb828).

Files with missing lines	Patch %	Lines
cnvlib/reports.py	90.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1178      +/-   ##
==========================================
+ Coverage   76.25%   76.39%   +0.13%     
==========================================
  Files          75       75              
  Lines        8428     8418      -10     
  Branches     1492     1487       -5     
==========================================
+ Hits         6427     6431       +4     
+ Misses       1568     1556      -12     
+ Partials      433      431       -2     

Flag	Coverage Δ
unittests	76.39% <95.45%> (+0.13%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.