[alisman]

Summary

This adds a new view kind, mutcompare, laid out as cBioPortal's group-comparison Mutations tab, and wires it into /portal/comparison right after Genomic Alterations, where the portal puts it.

The plot

• One gene's mutations in exactly two groups, mirrored about the protein.
  • Group A is drawn above the protein, group B below it.
  • The codon axis is on top and the Pfam domains sit in the middle.
  • Each half's y axis is labelled 0 and its max, the same way the portal labels it.
• Stems, colours, labels and domains come from the existing lib/lollipop.ts, the same code the results page's lollipop uses. There is no second lollipop implementation.
• The Y-Axis % / # switch: % is a share of the group's patients profiled for the gene (the portal's default); # is a count of patients. Both halves share one scale.
• Each group has a Driver / VUS badge selector beside the plot.
• Clicking a stem selects its codon and filters the table to it. Shift-click adds more codons; clicking the background clears the selection.

The gene picker

• A gene field, plus pills for the Highest Frequency: ten.
• The ranking follows the portal's genesSortedByMutationFrequency: genes are ordered by their higher share of mutated samples over profiled samples. That is frequency, not significance. The task brief said "ranked by significance", but the portal ranks by frequency, so this follows the portal.

Below the plot

• The Fisher line for the whole gene.
• A table with a row per protein change:
  • Protein Change, Annotation (OncoKB), Mutation Type
  • (A) and (B): patients (share %)
  • Log2 Ratio, shown as >10 / <-10 when either share is 0
  • p-Value: two-sided Fisher exact test
  • q-Value: Benjamini–Hochberg (BH)
  • Enriched in: a tie goes to B; the cell is coloured when q < 0.05
  • The header tips are the portal's, word for word. The table has the "Enriched in …" filter and "Significant only", and is sorted by p.
• Other than two groups is refused with the portal's message. The page leaves the tab out and says why, as it already does for the expression tabs, for other than two groups or a study with no mutation profile.

Other changes

• The axis, the badges and the codon selection are applied in the browser from the whole result (dataPart), so none of them re-runs the view. Changing the gene does re-run it.
• The per-gene mutated/profiled SQL that enrichment used moved, unchanged, from lib/views/enrichment/sql.ts to lib/federation-views.ts. Kinds can't import each other, so this lets both kinds share one copy.
• docs/PORTAL-PARITY.md: ticks this gap, with the leftovers as sub-items. It also ticks Per-sample CN segments (Copy-number segments view: per-sample segment tracks (the portal's CN Segments tab) #161), which is merged; its two sub-items stay unticked.

Legacy reference

Studied:

• GroupComparisonMutationsTab.tsx, GroupComparisonMutationMapper(Wrapper).tsx, GroupComparisonMutationTable.tsx
• LollipopGeneSelector.tsx, LollipopTooltipCountInfo.tsx, FisherExactTwoSidedTestLabel.tsx
• GroupComparisonUtils (getProteinChangeToMutationRowData, the MUTATIONS_* messages)
• In GroupComparisonStore / ComparisonStore: genesSortedByMutationFrequency, groupToProfiledPatients, showMutationsTab
• In react-mutation-mapper: LollipopPlotNoTooltip (the two-group layout) and AxisScaleSwitch

Matched:

• Shares are per profiled patient.
• A table count is the distinct patients with that protein change.
• A stem's height is the distinct (patient, change) pairs at that codon.
• Fisher's test is per change and BH is across the changes.
• Log ratio and Enriched-in follow the portal's rules, and the column tips are the portal's.
• The axis default is %, with a minimum max of 5 on #.
• The tooltip's percent uses the portal's format.
• The heading is "GENE mutations: (A) x vs (B) y".
• The Fisher line uses the portal's wording.

Deliberately different (cBioLab conventions):

• The page lists a left-out tab, with its reason, rather than showing an alert inside the tab.
• The caveat states the denominators and the counting rules.

Verification

Cross-checked against cbioportal.org's REST API on brca_tcga_pan_can_atlas_2018, comparing PAM50 basal-like (SUBTYPE=BRCA_Basal) against luminal A (BRCA_LumA). Data came from /clinical-data, /molecular-profiles/…_mutations/mutations/fetch and /sample-lists/…_sequenced, with scipy for the tests. Everything below is identical:

	cBioLab	cbioportal.org
TP53 profiled patients	171 / 499	171 / 499
TP53 mutated patients	153 / 53	153 / 53
TP53 protein changes	132	132
R342*	6 / 0, p 2.587e-4	6 / 0, p 2.587e-4
R175H	9 / 4, p 1.036e-3	9 / 4, p 1.036e-3
R273H	7 / 2, p 1.422e-3	7 / 2, p 1.422e-3
R196*	6 / 1, p 1.425e-3	6 / 1, p 1.425e-3
…the rest of the top 10	same	same
Ranking: PIK3CA (luminal A)	47.49%	47.49% (237/499)
Ranking: TTN (basal-like)	23.98%	23.98% (41/171)

Tests and checks:

• npm test passes (3,290 tests, 0 fail).
• New tests: test/mutcompare.test.ts, plus cases in test/comparison-layout.test.ts.
• node --test test/view-snapshots*.test.ts: a new mutcompare case, with answers recorded from the public warehouse. No outside HTTP calls needed recording: TP53's Genome Nexus answers were already recorded.
• test/boundary.test.ts covers the new mutcompareProfiledSql.
• npm run report:coverage && npm run check:coverage passes: 100% of changed lib lines run.
• npm run docs:gen and npm run docs:counts: only my example was added; no existing count moved.
• eslint and knip are clean. ruff did not run locally (no Python venv), and nothing in Python changed.
• I did not run the e2e gate locally. CI runs it.

Looked at in a browser (Playwright, dev server) with no console errors:

• The TP53 tab.
• Clicking the R175 stem: the table is filtered to R175H and the banner shows "Remove filter".
• The PIK3CA pill together with the # axis: H1047R is 6 vs 87, and the axis max is 98.

Not done

These are listed as unticked sub-items in PORTAL-PARITY.md:

• The Top / Bottom Y-Axis Max sliders.
• The lollipop's tracks (hotspots, OncoKB, PTM, exons, topology, PDB), its Legend toggle, and its SVG/PNG download. The table's copy and download.
• The table's Co-occurrence Pattern column (a mini oncoprint, which needs patient-level data), and the Custom Driver, Custom Driver Tier and MS columns.
• On the portal each group's badges filter that group alone. Here a badge hides that class in both groups.
• The Fisher line covers all of the gene's mutations only. The portal also gives it for filtered or selected mutations, which needs distinct patients over the selected changes.
• A stem's tooltip counts the stem's (patient, change) pairs, not distinct patients, and it doesn't show the other group's stem at the same codon.
• The gene picker is a field plus ten pills, not the portal's searchable 200-gene dropdown (the result already carries the 200).
• The ranking ignores the page's "OncoKB drivers only" setting and the event-type settings.
• The overlap banner.
• There is no browser check yet: the kind is e2ePending.

Preview

• /portal/comparison?doc=eyJzdHVkeSI6ICJicmNhX3RjZ2FfcGFuX2Nhbl9hdGxhc18yMDE4IiwgInZpZXdzIjogW10sICJjb2hvcnRzIjogeyJCYXNhbC1saWtlIjogeyJhbGwiOiBbeyJraW5kIjogImNsaW5pY2FsIiwgImF0dHJpYnV0ZSI6ICJTVUJUWVBFIiwgInZhbHVlcyI6IFsiQlJDQV9CYXNhbCJdfV19LCAiTHVtaW5hbCBBIjogeyJhbGwiOiBbeyJraW5kIjogImNsaW5pY2FsIiwgImF0dHJpYnV0ZSI6ICJTVUJUWVBFIiwgInZhbHVlcyI6IFsiQlJDQV9MdW1BIl19XX19fQ&tab=Mutations

🤖 Generated with Claude Code

https://claude.ai/code/session_01Q3ezTWaXhwWhadqM9gbFNK

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
cbiolab-genie	Ready	Preview	Nov 9, 2026 11:06pm UTC
cbiolab-new	Ready	Preview	Nov 9, 2026 11:06pm UTC
cohort-lab	Ready	Preview	Nov 9, 2026 11:06pm UTC