P. putida Gene Annotation Review Project

Overview

Systematic AI-assisted review of GO annotations for Pseudomonas putida, focusing primarily on the well-characterized KT2440 strain (UniProt: PSEPK). P. putida is a metabolically versatile soil bacterium of significant interest for bioremediation, industrial biotechnology, and plant growth promotion. Its annotations are predominantly from automated pipelines (InterPro2GO, UniProtKB-KW, TreeGrafter), with very few experimental annotations (~50 from PMIDs), making careful review essential.

Organism Details

• Species: Pseudomonas putida
• Primary strain: KT2440 (UniProt taxon code: PSEPK)
• Other strains: PSEPU (general P. putida)
• Genome: ~6.2 Mb, ~5,350 protein-coding genes
• Key biology: Aromatic compound degradation, solvent tolerance, plant root colonization, polyhydroxyalkanoate (PHA) biosynthesis, rare earth element utilization

Genome-wide expansion

The next phase expands the project from selected genes to a pathway-by-pathway
review of the KT2440 reference proteome. The working pattern is manual pathway
hole filling: start with a curated module, ask which steps are satisfiable in
PSEPK from lightweight UniProt metadata, then review only the genes needed to
resolve missing, ambiguous, over-propagated, or biologically important steps.

The detailed workplan is in
P_PUTIDA/P_PUTIDA_MODULE_PLAN.md. It covers
the module order, triage rules, module-editing expectations, and the first-pass
metadata snapshot. This phase deliberately separates metadata discovery from
full gene-review seeding: the first pass uses UniProt REST TSV metadata only,
not full UniProt flat files, GOA downloads, PMID caches, or review YAML stubs.

Lightweight data products:

• projects/P_PUTIDA/fetch_uniprot_metadata.py downloads basic UniProt metadata
  for proteome UP000000556.
• projects/P_PUTIDA/data/psepk_uniprot_metadata.tsv is the current metadata
  snapshot for triage and module clustering.
• projects/P_PUTIDA/data/psepk_uniprot_metadata.manifest.txt records the
  query, fields, source URL, and retrieval timestamp.
• projects/P_PUTIDA/build_gene_list.py derives a curator-facing gene list from
  the metadata snapshot.
• projects/P_PUTIDA/data/psepk_gene_list.tsv is the current whole-proteome
  gene list for module bucketing and fetch planning.
• projects/P_PUTIDA/partition_pathways.py joins the gene list to KEGG ppu
  pathway mappings and assigns each gene to one primary pathway/module/unknown
  bucket.
• projects/P_PUTIDA/data/psepk_pathway_partition.tsv is the current primary
  partition, one row per gene.
• projects/P_PUTIDA/data/psepk_pathway_membership.tsv preserves overlapping
  KEGG and UniPathway memberships.
• projects/P_PUTIDA/data/psepk_pathway_buckets.tsv summarizes the current
  pathway/module buckets.
• projects/P_PUTIDA/data/psepk_unknown_bucket.tsv contains genes with unknown
  function or minimal pathway signal.
• projects/P_PUTIDA/data/psepk_orphan_bucket.tsv contains EC/domain-family
  orphans that have some annotation signal but no current pathway bucket.
• projects/P_PUTIDA/build_pathway_worklist.py builds the PR-oriented pathway
  queue from the bucket table.
• projects/P_PUTIDA/data/psepk_pathway_worklist.tsv tracks module mapping,
  deep-research status, gene review coverage, provider coverage, and PR status
  per pathway/module bucket.
• projects/P_PUTIDA/extract_pathway_batch.py extracts a per-pathway gene
  checklist from KEGG/UniPathway membership.
• projects/P_PUTIDA/batches/ppu00400_tryptophan_biosynthesis.md is the first
  pilot pathway checklist.

While the Edison-backed Falcon route is unavailable, use OpenScientist for
gene-level research with a full two-hour provider allowance:

just deep-research-openscientist PSEPK <gene> --timeout 8100 \
  --extra-args --param timeout=7200 --param max_iterations=3

Use the same provider for reusable module-level research:

just module-deep-research-openscientist <module> --timeout 8100 \
  --extra-args --param timeout=7200 --param max_iterations=3

For PSEPK-specific pathway/module satisfiability research, use the taxon-aware
wrapper so the prompt includes the species constraint and local candidate genes
from the pathway partition:

just module-pathway-deep-research openscientist "<module or module title>" \
  <pathway-or-bucket> PSEPK --project P_PUTIDA --timeout 8100 \
  --extra-args --param timeout=7200 --param max_iterations=3

PaperBLAST can still be used opportunistically for protein-specific literature
lookup, but it is not the main workflow because the current wrapper is
Cloudflare-sensitive in this environment. Asta remains a lightweight gene-level
fallback; Falcon remains preferred for module synthesis when Edison is
available again. Use perplexity-lite only as a secondary fallback.

OpenScientist jobs commonly take more than 20 minutes and difficult runs can
approach the two-hour provider timeout. Do not use a 180-second smoke-test
timeout or infer failure from a quiet wrapper.

Completed Reviews

PSEPK (P. putida KT2440) — 18 genes reviewed

PSEPU (general P. putida) — 1 gene

Batch 2 — Selected for Review (50 new genes)

Fifty additional well-characterized KT2440 (taxon 160488) genes selected to broaden
functional coverage beyond the initial aromatic-catabolism/biotechnology focus, into
central carbon metabolism, stress response, DNA repair, motility, and membrane transport.
All were seeded via fetch-gene (UniProt + GOA). None overlap with previously present
gene folders.

Status: reviews complete. Each gene has a Falcon (Edison Scientific) deep-research
report (*-deep-research-falcon.md) and a fully reviewed *-ai-review.yaml — all GOA
annotations adjudicated (no PENDING), with descriptions, core_functions, and
references. The review pass corrected a number of mis-annotations and citation errors
surfaced in the raw reports (e.g. wrong PMIDs in exbB/fur/relA/pilA/hpd; cofactor terms
in fur (Zn→Fe(II)) and icd (NADP-specific); TreeGrafter/InterPro2GO over-propagations in
mdh, sdhA, groES, hfq, pvdQ; sigma-factor GO convention for rpoD/rpoH).

Stress response, chaperones & global regulation (12)

DNA repair & recombination (7)

Central carbon metabolism — TCA cycle (8)

Glycolysis / ED / anaplerotic & overflow metabolism (9)

Chemotaxis & motility (4)

Membrane transport, iron & nitrogen acquisition (8)

Aromatic / aromatic-amino-acid catabolism (2)

Batch 3 — Aromatic Amino Acid (AAA) Biosynthesis Pathway (16 genes)

The shikimate → chorismate → Trp/Phe/Tyr biosynthetic pathway in KT2440 (taxon
160488). Complements the existing aromatic catabolism coverage (ben/cat/pca) with the
anabolic route to aromatic amino acids. Seeded via fetch-gene; Falcon (Edison)
deep-research reports generated per gene. Status: reviews complete — all
annotations adjudicated, descriptions/core_functions/references populated (e.g. trpC and
aroQ over-propagation fixes; aroA bifunctional EPSPS/TyrA module captured). Name variants not resolving as primary UniProt
symbols in KT2440 are covered by a resolved paralog (aroD→aroQ, aroF/aroG→aroH, aroL→aroK)
or are differently named / fused in Pseudomonas (trpG, pheC, tyrA).

Shikimate trunk (common pathway, 7)

Tryptophan branch (6)

Phe/Tyr branch & aromatic aminotransferases (3)

Priority Genes for Future Review

Aromatic Catabolism (core P. putida biology)

• benABCD — Benzoate dioxygenase complex
• catABC — Catechol branch of β-ketoadipate pathway
• pcaGHBDCIJF — Protocatechuate branch
• nahABCDEF — Naphthalene degradation (if present in KT2440 derivatives)
• xylXYZ — Toluene/xylene degradation (TOL plasmid genes)

Central Metabolism & Biotechnology

• phaABCZ — Polyhydroxyalkanoate biosynthesis
• glk, zwf, edd, eda — Glucose catabolism (ED pathway, no EMP)
• gcd — Glucose dehydrogenase (periplasmic oxidation)
• PP_1084 (oleC), PP_1083 (oleD) — Olefin biosynthesis

Rare Earth Element Biology

• pedE — Ca²⁺-dependent ethanol dehydrogenase (counterpart to pedH)
• lanM — Lanmodulin (lanthanide-binding protein)
• lutH/lutABCDEF — Lanthanide uptake and transport

Solvent Tolerance

• ttgABC, ttgDEF, ttgGHI — Toluene efflux pumps (RND family)
• srpABC — Solvent resistance regulon

Plant Interactions

• pvdABCDES — Pyoverdine siderophore biosynthesis
• iaaM/iaaH — Indole-3-acetic acid biosynthesis (auxin)
• algABCDEFG — Alginate biosynthesis

Stress Response & Regulation

• rpoS — Stationary phase sigma factor
• gacA/gacS — Global regulatory two-component system
• fleQ — Flagellar/biofilm master regulator
• cbrA/cbrB — Carbon-nitrogen balance regulation

Related Projects

• SPKW-PSEPK — Analysis of UniProtKB-KW annotation patterns in P. putida
• REE — Rare earth element biology (pedH is a key gene)
• BIOSENSORS — Biosensor applications (BenR is relevant)

Notes

• P. putida KT2440 uses the Entner-Doudoroff pathway exclusively for glucose catabolism (no Embden-Meyerhof-Parnas pathway). This is important for annotation review — glycolysis terms may need careful handling.
• Many genes are annotated by homology to P. aeruginosa (PSEAE), but functional divergence is common — P. putida is non-pathogenic while P. aeruginosa is an opportunistic pathogen. Virulence-associated annotations transferred by homology should be scrutinized.
• The REE/lanthanide biology is relatively recent science (post-2011). Annotations for pedH and related genes may be incomplete or missing entirely.

STATUS

Genome-wide pathway/module curation

• [x] Re-scope P_PUTIDA from selected-gene project to genome-wide pathway/module curation umbrella.
• [x] Confirm working species/proteome: PSEPK / P. putida KT2440 / UniProt proteome UP000000556.
• [x] Confirm research provider policy: Asta for gene-level first-pass research; Falcon for module-level research; PaperBLAST optional; perplexity-lite secondary fallback.
• [x] Apply the 2026-07-20 operational override: OpenScientist for gene and module research while Edison/Falcon is unavailable, with full long-run timeouts.
• [x] Create module-first workplan in projects/P_PUTIDA/P_PUTIDA_MODULE_PLAN.md.
• [x] Refresh lightweight UniProt metadata snapshot before module clustering.
• [x] Build whole-proteome gene list from UniProt metadata.
• [x] Cluster all PSEPK proteins into first-pass module buckets from EC, KEGG, BioCyc, UniPathway, GO, InterPro/Pfam/PANTHER, keywords, and protein names.
• [x] Add species-aware module/pathway deep-research wrapper for module satisfiability work.
• [x] Select first pilot module/pathway batch: KEGG ppu00400 with tryptophan_biosynthesis as the seeded neutral module.
• [x] Complete Falcon module-level research for tryptophan_biosynthesis.
• [x] Complete Falcon module + pathway + PSEPK research for ppu00400.
• [x] Fetch, run Asta, curate, and validate the first pilot gene batch: 28/28 KEGG ppu00400 members.
• [x] Open the first module/pathway PR for ppu00400 / tryptophan_biosynthesis: PR #1874.
• [ ] For each later module batch, full fetch-gene only the genes selected by module review.
• [ ] Track module satisfiability gaps, over-annotations, missing GO terms, and candidate new module documents.

NOTES

2026-07-05

Started the genome-wide expansion plan. The curation unit is now a pathway/module
rather than an individual gene: use UniProt metadata to build the whole-proteome
map, use modules as expected pathway structure, and spend full gene-review effort
only where the module pass creates an actionable curation question.

Updated provider policy: Asta is the normal gene-level first-pass research
provider for this organism; Falcon is the preferred module-level provider.
PaperBLAST remains useful as an opportunistic lookup, but not the main workflow.

Fetched the UniProt-derived gene list for the KT2440 reference proteome:
projects/P_PUTIDA/data/psepk_gene_list.tsv has 5,527 protein records, all with
PP_ ordered locus tags. The snapshot contains 742 reviewed entries, 4,785
unreviewed entries, 1,452 entries with EC numbers, and 26 suggested review-name
collisions that are disambiguated by accession in the suggested_review_name
column.

Partitioned the 5,527 genes into first-pass pathway/module buckets using
organism-specific KEGG ppu pathway mappings as the primary source, UniPathway
as a fallback pathway source, and UniProt metadata heuristics for non-pathway
systems and unknowns. Current output has 161 primary buckets and 5,673
many-to-many pathway memberships. Primary assignments: 1,724 KEGG-pathway genes,
40 UniPathway-fallback genes, 1,789 functional-module genes, 1,149 orphan genes
with EC/domain signal, and 825 unknown/minimal-signal genes.

Added scripts/module_pathway_taxon_deep_research_wrapper.py and
templates/module_pathway_taxon_research.md.j2 for module + pathway + taxon
deep research. For PSEPK, the wrapper resolves projects/P_PUTIDA automatically,
uses the local pathway partition to add candidate genes, and writes reports under
projects/P_PUTIDA/deep-research/.

Built projects/P_PUTIDA/data/psepk_pathway_worklist.tsv with 161 pathway/module
bucket rows. Selected the first pilot batch: KEGG ppu00400 ("Phenylalanine,
tyrosine and tryptophan biosynthesis") seeded by the existing neutral
modules/tryptophan_biosynthesis.yaml module. The generated checklist
projects/P_PUTIDA/batches/ppu00400_tryptophan_biosynthesis.md has 28 candidate
KEGG members, 20 genes whose primary bucket is ppu00400, 15 existing review
files, and 0 existing Asta research files.

Attempted just module-deep-research-falcon tryptophan_biosynthesis, but the
command stalled in uv dependency/environment bootstrap before reaching
deep-research-client or the Falcon provider; the run was interrupted without
creating modules/tryptophan_biosynthesis-deep-research-falcon.md. The global
deep-research-client executable is not on PATH, so the configured execution
route remains blocked on uv run completing successfully.

Later in the same local session, fixed the provider execution route by updating
the deep-research wrappers to call uvx --from deep-research-client[cyberian]==0.2.7rc1 deep-research-client instead of bootstrapping the full project environment for
each run. Falcon is documented operationally as a long-running provider for this
project; the first completed Falcon module runs took about 22 minutes each.

Completed Falcon research for the pilot:

• modules/tryptophan_biosynthesis-deep-research-falcon.md
• projects/P_PUTIDA/deep-research/PSEPK__tryptophan_biosynthesis__ppu00400-deep-research-falcon.md

Fetched all missing review folders for the 28 KEGG ppu00400 members and ran
Asta gene-level retrieval for all 28. The regenerated checklist
projects/P_PUTIDA/batches/ppu00400_tryptophan_biosynthesis.md now records 28
existing review files, 28 curated review files, and 28 Asta reports.

Completed the first-pass curation of all ppu00400 batch gene reviews. Newly
curated stubs included the aroE paralogs, aroQ paralogs, aroF-I/aroF-II, pabA
(treated as the TrpG-like anthranilate synthase amidotransferase component for
the tryptophan module), quiA, amaC, PP_3768, and phhA. The Falcon
module/taxon report recommends keeping the neutral tryptophan_biosynthesis
module boundary strict at chorismate to L-tryptophan and treating upstream
shikimate genes, phenylalanine/tyrosine branch genes, and quinate-catabolic
genes as neighboring or shared-pathway context rather than core tryptophan
module members.

Validation run on 2026-07-05 PDT / 2026-07-06 UTC:

• just validate PSEPK <gene> for all 28 genes in
  projects/P_PUTIDA/batches/ppu00400_tryptophan_biosynthesis.tsv
• uv run linkml-validate -s src/ai_gene_review/schema/gene_review.yaml -C ModuleReview modules/tryptophan_biosynthesis.yaml

All validations passed. Remaining warnings are non-blocking and mostly note that
the gene reviews do not cite Asta reports directly; that is expected for this
light first pass because Asta is retained as retrieval context rather than used
as a hypothesis source unless it adds curation-changing evidence.

Opened draft PR #1874
for the ppu00400 / tryptophan_biosynthesis pilot batch.