figure-ed1-ed2-pipeline-seqlevel

Metadata

Description

## Description
Produce Extended Data ED1 (4 panels) and ED2 (4 panels). Implement directly — do not decompose further.

## File scope
- `paper_prep/figures/ed1/figure_ed1.{pdf,png,R/py}`, `caption.md`, `sources.tsv`
- `paper_prep/figures/ed2/figure_ed2.{pdf,png,R/py}`, `caption.md`, `sources.tsv`

## Figure spec (excerpt from MANUSCRIPT_SKELETON.md ED1, ED2)

### ED1 — Pipeline and per-arm flank inventory
| Panel | Content | Status | Source |
|---|---|---|---|
| ED1a | Pipeline schematic: 465 assemblies → 18,827 flanks → 15,668 PHRs → 15/50 communities | GENERATE | new schematic |
| ED1b | Per-arm flank counts (48 arms) with assembly QC overlay | GENERATE | `contig_classifications.tsv` — `SURVEY_01 §3` |
| ED1c | PHR length distribution (median 105 kb, mean 144 kb) | GENERATE | `all-vs-all.1Mb.p95.id95.len.tsv` — `SURVEY_01 §3` |
| ED1d | Chr18_q (NA18982#1) chimera evidence — wfmash + minimap2 dotplot + NNN gap + Flagger | GENERATE | `SURVEY_01 §1.5, §5 item 6` |

### ED2 — Sequence-level (50-community) detail
| Panel | Content | Status | Source |
|---|---|---|---|
| ED2a | UMAP / force-directed layout coloured by 50-community partition | READY/composite | `plot-seq-community-structure.R` outputs (`/moosefs/.../similarity/`) |
| ED2b | Within-community Jaccard distance bimodality (C1, C2, C3, C5, C6, C7, C11, C12) | GENERATE | `similarity.tsv.gz` per community subsets — `SURVEY_04 §1.10, §6 F10` |
| ED2c | Cross-arm affinity circular plot — 41 arms with edges weighted by absorbed sequences | GENERATE | `cross_arm_affinity_sequences.tsv` — `SURVEY_01 §6 F5` |
| ED2d | Confusion matrix Arm-Leiden vs Sequence-Leiden (15 × 50; ARI 0.35, NMI 0.76) | GENERATE | `arm-leiden` vs `seq-leiden` assignment TSVs |

## Validation
- [ ] All 8 panels (4 in ED1 + 4 in ED2); sources.tsv per ED
- [ ] Captions ≤ 200 words each; ≥ 2 metrics with TSV paths
- [ ] PDF + PNG per ED

## Inputs
- `paper_prep/synthesis/MANUSCRIPT_SKELETON.md`
- `paper_prep/surveys/SURVEY_01_pipeline.md`, `SURVEY_04_heterogeneity.md`

Depends on

• done synthesize-paper-rough
• done .assign-figure-ed1-ed2-pipeline-seqlevel

Required by

• done .flip-figure-ed1-ed2-pipeline-seqlevel
• done compile-manuscript-draft

Log

• 2026-05-05T05:28:12.319589466+00:00 Spawned by coordinator --executor claude --model opus
• 2026-05-05T05:28:33.524769367+00:00 Starting: ED1+ED2 figure generation. Inspecting inputs and data sources.
• 2026-05-05T05:31:16.606236699+00:00 Plan: 8 panels using R/ggplot2 (no patchwork — using grid.layout for composition). Data sources confirmed: ED1a: pipeline schematic (synthetic, drawn in R/grid) ED1b: per-arm flank counts from all-vs-all.1Mb.p95.id95.len.tsv (18,827) + contig_classifications.tsv QC overlay ED1c: PHR length distribution from all-vs-all.1Mb.p95.id95.len.tsv (signal-bearing rows; median 105 kb) ED1d: chr18_q chimera schematic (NA18982#1 JBKABS010000018.1 ~83.37-83.38 Mb, NNN gap, TTAGGG tract) ED2a: 50-community 'UMAP' — using MDS on arm-pair distances or PCA on similarity since umap pkg unavailable; will load existing dist_matrix.tsv ED2b: within-community Jaccard from similarity.tsv.gz streamed-filter for C1,C2,C3,C5,C6,C7,C11,C12 (using seq-leiden assignments) ED2c: chord-style cross-arm affinity (no circlize — will draw using base R plot + arcs) from cross_arm_affinity_sequences.tsv ED2d: confusion matrix arm-leiden (15) x seq-leiden (50)
• 2026-05-05T05:46:33.459235447+00:00 Committed: 74838fa — pushed to origin/wg/agent-696/figure-ed1-ed2-pipeline-seqlevel. 12 files: ED1 + ED2 PDFs/PNGs/R scripts, captions, sources.tsv, helper script.
• 2026-05-05T05:51:18.959977911+00:00 Validated: ED1+ED2 each have 4 panels (PDF+PNG); per-ED captions ≤ 200 word body (192 / 196); per-ED sources.tsv with ≥ 2 metric+TSV-path entries; ARI 0.35 / NMI 0.76 match SURVEY_01 §1.11.
• 2026-05-05T05:51:22.419008778+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
• 2026-05-05T06:02:16.504780981+00:00 PendingEval → Done (evaluator passed; downstream unblocks)