feat: integrate issue #32 field-gap additions (papers, datasets, software, databases, resources)

[benjibromberg]

Summary

Integrates the 92 field-gap candidate additions from the issue #32 analysis into the CAAIL canonical files, following the zotero-to-caail-sync workflow with mandatory adversarial reviewer subagents. The branch was rebased onto post-#33 main and reconciled against the new Taxonomy.md-linked matrix.

Closes #32.

What's included (by category)

File(s)	Added	Notes
Papers.md	36 papers — 27 primary refs (#200–225, #235) + 9 reviews (#226–234)	matrix cells + reference entries in lockstep
Datasets/	16 atlases / FM corpora / bioprocess + benchmark deposits	across Sheep, Duck, Fish, Crustacean, Mollusk, Human/CHO reference, Benchmarks
Software.md	16 open-source tools	protein structure/design, metabolic modeling, multi-omics
Databases.md	11 repositories	incl. a new Glycomics & Glycoprotein section
OtherResources.md + Primers/	14 initiatives, courses, bibliographies, primers

Corpus counts: papers 195 → 231, software 70 → 86, databases 72 → 83, datasets 130 → 146.

Rebase onto post-#33 main

PR #33 merged after this branch was cut and rewrote the Papers.md matrix (Taxonomy-linked axes, the Bioprocess control → Bioprocess & Scale-Up rename, a new Chemometrics row, and reference IDs #198/#199). Reconciliation:

• Renumbered 198–233 → 200–235 to avoid colliding with feat: matrix classification audit, Taxonomy, and Papers Explorer upgrades #33's new #198 (Thévenot) / #199 (Rohart). Citation text is unchanged from the prior caail-citation-reviewer pass — only the <a id> numbers and matrix (#N) anchors shifted.
• Re-applied every matrix anchor into feat: matrix classification audit, Taxonomy, and Papers Explorer upgrades #33's reclassified cells (append, not overwrite) using the renamed Bioprocess & Scale-Up column and Taxonomy axis links (no Wikipedia links reintroduced).
• Two empty cells filled: Reinforcement Learning × Bioprocess & Scale-Up (#200–203), CNN × Cellular Engineering (#218).

Classification re-grounding (caail-classification-reviewer)

Two placements were re-verified against the papers' own methods under the stricter post-#33 taxonomy:

• Xu et al. 2025 (#208) — comparative Raman/capacitance viability-monitoring study; multi-anchored SVM + Ensemble Learning + Chemometrics × Bioprocess & Scale-Up (SVR and PLS are both genuinely implemented and competitive, per the Roell #32 multi-anchor precedent).
• Serpe et al. 2025 (#217) — corrected from CNN × Scaffolding to Ensemble Learning × Scaffolding: the imaging ML is a Trainable-Weka Fast Random Forest pixel classifier, not a CNN (the only CNN mention is a related-work citation). Column stays Scaffolding (the deliverable is plant-scaffold quality assessment; no bioreactor was used).
• CellFM (#235) classified to Foundation Models: Masked Language Modeling × Cellular Engineering (masked gene-expression pretraining objective).

Curation deviations from the raw issue list (carried from the #32 checkpoints)

• g2f held out of Software.md — repo is dead and the CRAN package is archived (no live canonical home).
• ProteomeXchange promoted from an inline PRIDE mention to a dedicated Databases.md entry, placed beside PRIDE rather than the issue's "spectral databases" section (which is reserved for spectral libraries).
• Reviewer-flagged source corrections applied (oyster species → Crassostrea hongkongensis; dead Tabula Sapiens / scBaseCount URLs repointed; GlyGen attribution; unsupported counts/superlatives removed). The BioML-bench companion in Datasets/Benchmarks.md points at Papers.md #225.

Verification

• pnpm --dir site lint:papers → 0 errors (matrix ↔ reference integrity OK).
• pnpm --dir site build → succeeds (Taxonomy build guard + count-drift guards pass).
• pnpm --dir site test → 296 passed; pnpm --dir site test:e2e → 34 passed (parser + e2e ground-truth synced to the new counts).

Every drafted entry was verified before commit by the read-only adversarial reviewer subagents (caail-citation-reviewer, caail-claim-reviewer, caail-classification-reviewer); the agent that wrote an entry never reviewed it.

Update — 4 additional new-since-#32 Zotero items

A follow-up Zotero ⇄ repo sync pass appended four resources added to the caail Zotero library after the #32 set above (Papers IDs continue from 236):

• feat(papers) — Tac, Gardner & Kuhl 2026, Generative AI creates delicious, sustainable, and nutritious burgers (arXiv 2602.03092), ref #236, anchored in Deep Learning × Sensory Prediction. The model is a multinomial-diffusion + score-based generative architecture; the classification reviewer flagged a taxonomy gap (no diffusion/score-based row exists) and placed it in the Deep Learning catch-all as the best available row — worth considering a dedicated "Diffusion & Score-Based Generative Models" row later.

• feat(data) — CRISPR/Cas9 GGTA1-knockout bovine satellite-cell RNA-seq deposit (GEO GSE330550; D'Costa et al. 2026) → Datasets/Cow.md. Dataset-only (the paper applies CRISPR, not an AI/ML method, so no matrix entry); the thin grkenney/AGAL R package was left out.

• feat(software) — Context7 (upstash/context7, MIT) → Software.md § AI Agents & Foundation Models, framed honestly as general AI-agent developer infrastructure (not cell-ag-specific).

• feat(databases) — Pando (Foray Bioscience) → Databases.md § Ecosystem & Industry Directories, a commercial in vitro plant knowledge base (plant-side cell-ag); the "largest" claim is attributed to Foray.

• feat(data) — hyperlinked 115 bare accession IDs (GSE / PRJNA / SRA / CRA / PRJCA / PXD / CNP / GVM / OMIX / JR…) across the per-species Datasets/*.md pages to their canonical registries (GitHub + site), skipping already-linked ones.

Counts re-synced: papers 231→232, software 86→87, databases 83→84, datasets 146→147. lint:papers (0 errors), build, 296 vitest, and 34 e2e all pass. Each entry was verified by the read-only reviewer subagents before commit.

🤖 Generated with Claude Code