Alzheimer Disease Gene Review Project

Overview

This project reviews human GO annotations for genes central to Alzheimer disease
genetics and biology. The scope combines Mendelian/familial Alzheimer disease
genes, high-confidence common and rare variant risk genes, and pathway genes
needed to curate reusable disease-relevant modules.

The initial project focus is human Alzheimer disease (MONDO:0004975) with
module curation in the root-level modules/ directory.

Model Species

Primary: Homo sapiens (human)

• UniProt species code: HUMAN
• Review scope: human gene annotation reviews plus pathway/module curation
• Initial gene set: bounded seed list for review; not every included gene is
  asserted to be independently causal for Alzheimer disease

Genes for Review

Priority 1: Foundational Alzheimer Genetics

Priority 2: Amyloid Processing and Tau Biology

Priority 3: Lipid, Endocytosis, Complement, and Microglial Risk Network

Pathway Modules

Curate disease-relevant pathway and complex models as ModuleReview YAML files
under modules/:

Rendered module page: pages/modules/alzheimer_disease_pathways.html.

Validate each module with:

uv run linkml-validate -s src/ai_gene_review/schema/gene_review.yaml -C ModuleReview modules/<module>.yaml

Source Anchors

• Open Targets Platform query for Alzheimer disease (MONDO_0004975), accessed
  2026-06-19. Top associated targets included APP, PSEN1, PSEN2, APOE, SORL1,
  ADAM10, TREM2, ABCA7, CLU, BIN1, PLCG2, and APH1B.
• GWAS Catalog REST query for efo_trait=Alzheimer disease, accessed
  2026-06-19, used to confirm the disease trait mapping and GWAS context.
• Bellenguez et al. 2022, "New insights into the genetic etiology of Alzheimer's
  disease and related dementias" (PMID 35379992).
• Kunkle et al. 2019, "Genetic meta-analysis of diagnosed Alzheimer's disease
  identifies new risk loci and implicates amyloid beta, tau, immunity and lipid
  processing" (PMID 30820047).
• Sims et al. 2017, "Rare coding variants in PLCG2, ABI3, and TREM2 implicate
  microglial-mediated innate immunity in Alzheimer's disease"
  (PMID 28714976).

STATUS

• [x] Create project scope and seed gene list
• [x] Fetch UniProt, GOA, and publication caches for all frontmatter genes
• [x] Run deep research or record manual notes for Priority 1 genes
• [x] Complete Priority 1 gene annotation reviews
• [x] Draft and validate Alzheimer pathway modules under modules/
• [x] Review Priority 2 and Priority 3 genes
• [x] Render project page and update project index

NOTES

2026-06-19

• Created project scaffold with 34 human seed genes.
• Seed list is intentionally mixed: Mendelian AD genes, late-onset AD risk genes,
  and mechanistic pathway genes required for module curation.
• Open Targets and GWAS Catalog were used as scoping anchors; per-gene reviews
  still need the standard just fetch-gene human <GENE> and deep-research steps
  before curation decisions.
• Fetched and seeded missing Priority 1 gene data for PSEN1, PSEN2, APOE, TREM2,
  SORL1, ABCA7, and ADAM10. Publication-cache confirmation is complete for these
  genes: PSEN1 57/57, PSEN2 18/18, APOE 104/104, TREM2 31/31, SORL1 34/34,
  ABCA7 7/7, and ADAM10 44/44. Falcon deep research for PSEN1 stalled without
  writing an artifact; direct perplexity-lite failed with a 401 insufficient-quota
  error, so manual notes were started in genes/human/PSEN1/PSEN1-notes.md.
• Completed PSEN1 first-pass review: all 197 seeded GO annotations have review
  actions, core gamma-secretase function is defined, knowledge gaps are in
  genes/human/PSEN1/PSEN1-notes.md, and just validate human PSEN1 passes
  cleanly. Action distribution: 65 ACCEPT, 52 KEEP_AS_NON_CORE, 57
  MARK_AS_OVER_ANNOTATED, 7 MODIFY, 16 UNDECIDED.
• Completed PSEN2 first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/PSEN2/PSEN2-notes.md from cached UniProt/GOA/publication
  evidence. All 79 seeded GO annotations have review actions, core PSEN2
  gamma-secretase function is defined with late endosome/lysosome emphasis, and
  just validate human PSEN2 passes cleanly. Action distribution: 40 ACCEPT, 24
  KEEP_AS_NON_CORE, 11 MARK_AS_OVER_ANNOTATED, 4 UNDECIDED.
• Completed APOE first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/APOE/APOE-notes.md from cached UniProt/GOA/publication evidence.
  All 293 seeded GO annotations have review actions, core APOE
  lipid-transfer/lipoprotein-clearance functions are defined, and just validate human APOE passes cleanly. Action distribution: 145 ACCEPT, 120
  KEEP_AS_NON_CORE, 28 MARK_AS_OVER_ANNOTATED.
• Completed TREM2 first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/TREM2/TREM2-notes.md from cached UniProt/GOA/publication
  evidence. All 273 seeded GO annotations have review actions, core TREM2
  lipid/apolipoprotein ligand-sensing and TYROBP-linked receptor signaling
  functions are defined, and just validate human TREM2 passes cleanly. Action
  distribution: 117 ACCEPT, 151 KEEP_AS_NON_CORE, 5 MARK_AS_OVER_ANNOTATED.
• Completed SORL1 first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/SORL1/SORL1-notes.md from cached UniProt/GOA/publication
  evidence. All 168 seeded GO annotations have review actions, core SORL1 cargo
  receptor, APP/amyloid-beta sorting, and Golgi/endosomal trafficking functions
  are defined, and just validate human SORL1 passes cleanly. Action
  distribution: 92 ACCEPT, 44 KEEP_AS_NON_CORE, 25 MARK_AS_OVER_ANNOTATED, 4
  MODIFY, 3 UNDECIDED.
• Completed ABCA7 first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/ABCA7/ABCA7-notes.md from cached UniProt/GOA/publication
  evidence. All 81 seeded GO annotations have review actions, core ABCA7
  ATP-coupled phospholipid transport, apoA-I phospholipid efflux, and
  phagocytic/amyloid-clearance functions are defined, and just validate human ABCA7 passes cleanly. Action distribution: 55 ACCEPT, 20 KEEP_AS_NON_CORE, 4
  MODIFY, 2 MARK_AS_OVER_ANNOTATED.
• Completed ADAM10 first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/ADAM10/ADAM10-notes.md from cached UniProt/GOA/publication
  evidence. All 151 seeded GO annotations have review actions, core ADAM10
  metalloendopeptidase/ectodomain-shedding, APP alpha-secretase, Notch, and
  regulated membrane-localization functions are defined, and just validate human ADAM10 passes cleanly. Action distribution: 88 ACCEPT, 42
  KEEP_AS_NON_CORE, 19 MARK_AS_OVER_ANNOTATED, 1 MODIFY, 1 REMOVE.
• Fetched and seeded amyloid/gamma-secretase pathway batch data for BACE1,
  NCSTN, APH1A, APH1B, and PSENEN. Seeded GOA annotation counts: BACE1 140,
  NCSTN 99, APH1A 73, APH1B 43, PSENEN 71.
• Completed BACE1 first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/BACE1/BACE1-notes.md from cached UniProt/GOA/publication
  evidence. All 140 seeded GO annotations have review actions, core BACE1
  beta-secretase/aspartyl endopeptidase, APP beta-cleavage, amyloid-beta
  formation, and TGN/endosomal trafficking functions are defined, and just validate human BACE1 passes cleanly. Action distribution: 94 ACCEPT, 24
  KEEP_AS_NON_CORE, 20 MARK_AS_OVER_ANNOTATED, 2 MODIFY.
• Completed NCSTN first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/NCSTN/NCSTN-notes.md from cached UniProt/GOA/publication
  evidence. All 99 seeded GO annotations have review actions, core nicastrin
  gamma-secretase complex adaptor/substrate-recruitment contribution to
  intramembrane aspartyl endopeptidase activity is defined, and just validate human NCSTN passes cleanly. Action distribution: 73 ACCEPT, 12
  KEEP_AS_NON_CORE, 10 MARK_AS_OVER_ANNOTATED, 4 MODIFY.
• Completed APH1A first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/APH1A/APH1A-notes.md from cached UniProt/GOA/publication
  evidence. All 73 seeded GO annotations have review actions, core APH1A
  gamma-secretase complex adaptor/assembly contribution to intramembrane
  aspartyl endopeptidase activity is defined, and just validate human APH1A
  passes cleanly. Action distribution: 50 ACCEPT, 12 KEEP_AS_NON_CORE, 9
  MARK_AS_OVER_ANNOTATED, 2 MODIFY.
• Completed APH1B first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/APH1B/APH1B-notes.md from cached UniProt/GOA/publication
  evidence. All 43 seeded GO annotations have review actions, core APH1B
  gamma-secretase complex adaptor/assembly contribution to intramembrane
  aspartyl endopeptidase activity is defined with APH1B-specific substrate
  preference questions, and just validate human APH1B passes cleanly. Action
  distribution: 37 ACCEPT, 4 MARK_AS_OVER_ANNOTATED, 1 MODIFY, 1
  KEEP_AS_NON_CORE.
• Completed PSENEN first-pass review: falcon deep research timed out after 180
  seconds without writing an artifact, so manual notes were created in
  genes/human/PSENEN/PSENEN-notes.md from cached UniProt/GOA/publication
  evidence. All 71 seeded GO annotations have review actions, core PEN-2
  gamma-secretase endopeptidase-activator contribution, presenilin
  endoproteolysis, complex stability, trafficking, and APP/Notch processing
  functions are defined, and just validate human PSENEN passes cleanly. Action
  distribution: 55 ACCEPT, 8 KEEP_AS_NON_CORE, 7 MARK_AS_OVER_ANNOTATED, 1
  MODIFY.
• Completed CLU second-pass cleanup: existing annotation reviews were already
  present, but status was still IN_PROGRESS and one repeated GO term had
  inconsistent actions. Reconciled negative regulation of protein-containing complex assembly to ACCEPT, added the available falcon deep-research file as
  supporting context, set status to COMPLETE, rendered the review, and just validate human CLU now passes cleanly. Action distribution: 73 ACCEPT, 56
  KEEP_AS_NON_CORE, 22 MARK_AS_OVER_ANNOTATED, 4 MODIFY.
• Completed BIN1 first-pass review: fetched gene data, GOA annotations,
  publications, and PANTHER family PTHR46514. Falcon deep research timed out
  after 180 seconds without writing an artifact, so manual notes were created in
  genes/human/BIN1/BIN1-notes.md from cached UniProt/GOA/PANTHER/publication
  evidence. All 112 seeded GO annotations have review actions, core BIN1
  BAR-domain phospholipid/membrane-curvature, lipid tubulation, endocytosis,
  clathrin/dynamin adaptor, synaptic vesicle, and T-tubule functions are
  defined, and just validate human BIN1 passes cleanly. Action distribution:
  62 ACCEPT, 31 KEEP_AS_NON_CORE, 18 MARK_AS_OVER_ANNOTATED, 1 MODIFY.
• Completed PICALM first-pass review: fetched gene data, GOA annotations,
  publications, and PANTHER family PTHR22951. Falcon deep research timed out
  after 180 seconds without writing an artifact, so manual notes were created in
  genes/human/PICALM/PICALM-notes.md from cached UniProt/GOA/Reactome/publication
  evidence. All 113 seeded GO annotations have review actions, core PICALM
  phosphoinositide-bound clathrin adaptor, coated-vesicle curvature/maturation,
  clathrin coat assembly, and small R-SNARE cargo-sorting functions are defined,
  and just validate human PICALM passes cleanly. Action distribution: 80
  ACCEPT, 22 KEEP_AS_NON_CORE, 10 MARK_AS_OVER_ANNOTATED, 1 MODIFY.
• Completed CD33 first-pass review: fetched gene data, GOA annotations,
  publications, and PANTHER family PTHR12035. Falcon deep research timed out
  after 180 seconds without writing an artifact, so manual notes were created in
  genes/human/CD33/CD33-notes.md from cached UniProt/GOA/Reactome/PANTHER/publication
  evidence. All 45 seeded GO annotations have review actions, core CD33/Siglec-3
  sialic-acid binding, cell-surface inhibitory receptor signaling, ITIM-linked
  SHP-1/SHP-2 recruitment, and negative myeloid/monocyte activation functions
  are defined, and just validate human CD33 passes cleanly. Action
  distribution: 29 ACCEPT, 5 KEEP_AS_NON_CORE, 8 MARK_AS_OVER_ANNOTATED, 2
  MODIFY, 1 UNDECIDED.
• Completed CR1 first-pass review: fetched gene data, GOA annotations,
  publications, and confirmed existing PANTHER family PTHR45656. Falcon deep
  research timed out after 180 seconds without writing an artifact, so manual
  notes were created in genes/human/CR1/CR1-notes.md from cached
  UniProt/GOA/Reactome/PANTHER/publication evidence. All 64 seeded GO
  annotations have review actions, core CR1/CD35 complement receptor activity,
  C3b/C4b binding, immune-adherence clearance, factor I cofactor/convertase
  decay, and negative complement-regulatory functions are defined, and just validate human CR1 passes cleanly. Action distribution: 42 ACCEPT, 17
  KEEP_AS_NON_CORE, 3 MARK_AS_OVER_ANNOTATED, 2 MODIFY.
• Completed CD2AP first-pass review: fetched gene data, GOA annotations,
  publications, and confirmed existing PANTHER family PTHR14167. Falcon deep
  research timed out after 180 seconds without writing an artifact, so manual
  notes were created in genes/human/CD2AP/CD2AP-notes.md from cached
  UniProt/GOA/PANTHER/publication evidence. All 72 seeded GO annotations have
  review actions, core CD2AP SH3-domain adaptor/scaffold, actin remodeling,
  junction/slit-diaphragm, and endosomal trafficking functions are defined, and
  just validate human CD2AP passes cleanly. Action distribution: 36 ACCEPT, 9
  KEEP_AS_NON_CORE, 19 MARK_AS_OVER_ANNOTATED, 7 MODIFY, 1 UNDECIDED.
• Completed INPP5D first-pass review: fetched gene data, GOA annotations,
  publications, and PANTHER family PTHR46051. Falcon deep research timed out
  after 180 seconds without writing an artifact, so manual notes were created in
  genes/human/INPP5D/INPP5D-notes.md from cached
  UniProt/GOA/Reactome/PANTHER/publication evidence. All 56 seeded GO
  annotations have review actions, plus one NEW phosphotyrosine-residue binding
  annotation was added to replace repeated generic protein-binding entries. Core
  SHIP1 PtdIns(3,4,5)P3/IP4 5-phosphatase, phosphotyrosine-motif recruitment,
  and immune receptor signal-attenuation functions are defined, and just validate human INPP5D passes cleanly. Action distribution: 34 ACCEPT, 4
  KEEP_AS_NON_CORE, 3 MARK_AS_OVER_ANNOTATED, 12 MODIFY, 2 UNDECIDED, 1 REMOVE,
  1 NEW.
• Completed PLCG2 first-pass review: fetched gene data, GOA annotations,
  publications, and PANTHER family PTHR10336. Falcon deep research timed out
  after 180 seconds without writing an artifact, so manual notes were created in
  genes/human/PLCG2/PLCG2-notes.md from cached
  UniProt/GOA/Reactome/PANTHER/publication evidence. All 167 seeded GO
  annotations have review actions, core PLCG2 PIP2 phospholipase, IP3/DAG and
  calcium signaling, phosphotyrosine-dependent recruitment, BCR/antigen/C-type
  lectin receptor signaling, and membrane/raft/ruffle localization functions
  are defined, and just validate human PLCG2 passes cleanly. Alzheimer-relevant
  TREM2/TLR microglial phenotypes were retained as important non-core cellular
  outcomes. Action distribution: 83 ACCEPT, 56 KEEP_AS_NON_CORE, 19 MODIFY, 7
  MARK_AS_OVER_ANNOTATED, 1 UNDECIDED, 1 REMOVE.
• Completed SPI1 first-pass review: fetched gene data, GOA annotations,
  publications, and PANTHER family PTHR11849. Falcon deep research timed out
  after 180 seconds without writing an artifact, so manual notes were created in
  genes/human/SPI1/SPI1-notes.md from cached
  UniProt/GOA/Reactome/PANTHER/publication evidence. Added the cached PU.1
  macrophage binding-site selection paper PMID:23658224 as an extra reference.
  All 140 seeded GO annotations have review actions, core PU.1 ETS/PU-box
  cis-regulatory DNA binding, RNA polymerase II transcription-factor activity,
  pioneer/chromatin accessibility control, and B-cell/myeloid hematopoietic
  lineage functions are defined, and just validate human SPI1 passes cleanly.
  Microglial/macrophage and immune-output terms were retained as non-core where
  they represent downstream cell-state consequences. Action distribution: 74
  ACCEPT, 32 KEEP_AS_NON_CORE, 21 MODIFY, 11 MARK_AS_OVER_ANNOTATED, 2
  UNDECIDED.
• Completed MS4A4A first-pass review: fetched gene data, GOA annotations,
  publications, and PANTHER family PTHR23320. Falcon deep research timed out
  after 180 seconds without writing an artifact, so manual notes were created in
  genes/human/MS4A4A/MS4A4A-notes.md from cached
  UniProt/GOA/PANTHER/publication evidence. All 9 seeded GO annotations have
  review actions, core four-pass MS4A4A membrane, plasma membrane raft, Golgi,
  and endoplasmic reticulum localization is defined, and just validate human MS4A4A passes cleanly. Generic interactome protein-binding annotations were
  marked over-annotated; TREM2/sTREM2 modulation was recorded as an
  Alzheimer-relevant unresolved biological question rather than an invented
  molecular function. Action distribution: 7 ACCEPT, 2
  MARK_AS_OVER_ANNOTATED.
• Completed MS4A6A first-pass review: fetched gene data, GOA annotations,
  publications, and confirmed PANTHER family PTHR23320. Falcon deep research
  timed out after 180 seconds without writing an artifact, so manual notes were
  created in genes/human/MS4A6A/MS4A6A-notes.md from cached
  UniProt/GOA/PANTHER/publication evidence. All 6 seeded GO annotations have
  review actions, core MS4A6A four-pass membrane/trans-Golgi localization is
  defined, and just validate human MS4A6A passes cleanly. The plasma-membrane
  IBA was modified to the directly supported trans-Golgi network term, while
  broad receptor-signaling and generic protein-binding annotations were marked
  over-annotated. Action distribution: 3 ACCEPT, 2
  MARK_AS_OVER_ANNOTATED, 1 MODIFY.
• Completed EPHA1 first-pass review: fetched gene data, GOA annotations,
  publications, Reactome records, and PANTHER family PTHR46877. Falcon deep
  research timed out after 180 seconds without writing an artifact, so manual
  notes were created in genes/human/EPHA1/EPHA1-notes.md from cached
  UniProt/GOA/Reactome/PANTHER/publication evidence. All 42 seeded GO
  annotations have review actions, core EPHA1 transmembrane ephrin receptor and
  receptor protein tyrosine kinase functions are defined, and just validate human EPHA1 passes cleanly. Context-dependent adhesion, migration,
  angiogenesis, fibronectin-binding, and kinase-binding outputs were kept as
  non-core; indirect stress-fiber assembly transfers were marked
  over-annotated. Action distribution: 30 ACCEPT, 10 KEEP_AS_NON_CORE, 2
  MARK_AS_OVER_ANNOTATED.
• Completed FERMT2 first-pass review: fetched gene data, GOA annotations,
  publications, Reactome records, and PANTHER family PTHR16160. Falcon deep
  research timed out after 180 seconds without writing an artifact, so manual
  notes were created in genes/human/FERMT2/FERMT2-notes.md from cached
  UniProt/GOA/Reactome/PANTHER/publication evidence. All 85 seeded GO
  annotations have review actions, core kindlin-2 integrin-binding/co-activator,
  phosphoinositide/PIP3 membrane-engagement, and actin/ILK-linked
  focal-adhesion scaffold functions are defined, and just validate human FERMT2 passes cleanly. Wnt, TGF-beta/Smad, EMT, endothelial barrier,
  trophoblast migration, and MSC fate outputs were kept as non-core; generic
  protein-binding and imprecise cell-surface terms were modified to more
  informative replacements. Action distribution: 52 ACCEPT, 30
  KEEP_AS_NON_CORE, 3 MODIFY.
• Completed CASS4 first-pass review: fetched gene data, GOA annotations,
  publications, and confirmed PANTHER family PTHR10654. Falcon deep research
  timed out after 180 seconds without writing an artifact, so manual notes were
  created in genes/human/CASS4/CASS4-notes.md from cached
  UniProt/GOA/PANTHER/publication evidence. All 15 seeded GO annotations have
  review actions, core CASS4/HEPL Cas-family focal-adhesion docking/scaffold,
  FAK/protein-tyrosine-kinase binding/regulation, cell spreading, and migration
  functions are defined, and just validate human CASS4 passes cleanly. The
  broad receptor-tyrosine-kinase signaling IBA was modified to
  integrin-mediated signaling; AKT pathway activation was kept as non-core NSCLC
  context. Action distribution: 13 ACCEPT, 1 KEEP_AS_NON_CORE, 1 MODIFY.
• Completed MAPT first-pass review: fetched gene data, 198 GOA annotations,
  publications, and confirmed PANTHER family PTHR11501. Falcon deep research
  timed out after 180 seconds without writing an artifact, so manual notes were
  created in genes/human/MAPT/MAPT-notes.md from cached
  UniProt/GOA/PANTHER/publication evidence. All seeded annotations have review
  actions, core tau microtubule-stabilizing/axon/membrane-cortex linker
  function and Alzheimer-relevant tau self-assembly/neurofibrillary-tangle
  pathology are defined, and just validate human MAPT passes cleanly. Synaptic,
  mitochondrial, secretion, glial, stress-response, nuclear, and partner-binding
  contexts were kept as non-core where credible; generic protein binding and
  broad nucleic-acid/process terms were marked over-annotated. The part_of tubulin complex annotation was modified to tubulin binding. Action
  distribution: 77 ACCEPT, 70 KEEP_AS_NON_CORE, 50 MARK_AS_OVER_ANNOTATED, 1
  MODIFY.
• Completed GSK3B first-pass review: fetched gene data, 390 GOA annotations,
  publications, Reactome-derived entries, and confirmed PANTHER family
  PTHR24057. Falcon deep research timed out after 180 seconds without writing
  an artifact, so manual notes were created in
  genes/human/GSK3B/GSK3B-notes.md from cached
  UniProt/GOA/Reactome/PANTHER/publication evidence. All seeded annotations have
  review actions, core GSK3B ATP-dependent protein serine/threonine kinase,
  tau-protein kinase, Wnt/beta-catenin destruction-complex, protein-turnover,
  and glycogen/insulin metabolic-regulation functions are defined, and just validate human GSK3B passes cleanly. Broad apoptosis, EMT, ER-stress, mTOR,
  cilia, circadian, immune, synaptic, developmental, and centrosomal outputs were
  kept as non-core; generic protein binding and broad signal-transduction
  terms were marked over-annotated; generic kinase terms were modified to
  protein serine/threonine kinase activity. Action distribution: 172 ACCEPT, 126
  KEEP_AS_NON_CORE, 84 MARK_AS_OVER_ANNOTATED, 8 MODIFY.
• Completed LRP1 first-pass review: fetched gene data, 124 GOA annotations,
  publications, and confirmed PANTHER family PTHR22722. Falcon deep research
  timed out after 180 seconds without writing an artifact, so manual notes were
  created in genes/human/LRP1/LRP1-notes.md from cached
  UniProt/GOA/PANTHER/publication evidence. All seeded annotations have review
  actions, core LRP1 LDLR-family cargo/scavenger receptor, apolipoprotein and
  lipoprotein receptor, receptor-mediated endocytosis/internalization, amyloid
  beta clearance/transcytosis, blood-brain-barrier transport, and tau/MAPT uptake
  context are defined, and just validate human LRP1 passes cleanly. Vascular,
  extracellular-matrix, inflammatory, infection/toxin, intracellular-domain, and
  signaling side branches were kept as non-core; generic protein/RNA/complex
  binding and broad signaling-receptor labels were marked over-annotated. Action
  distribution: 77 ACCEPT, 29 KEEP_AS_NON_CORE, 18 MARK_AS_OVER_ANNOTATED.
• Completed ABCA1 first-pass review: fetched gene data, 145 GOA annotations,
  publication caches, Reactome-derived entries, and confirmed the UniProt/PANTHER
  ABCA-family assignment PTHR19229:SF34. Falcon deep research timed out after
  180 seconds without writing an artifact, so manual notes were created in
  genes/human/ABCA1/ABCA1-notes.md from cached
  UniProt/GOA/Reactome/publication evidence. All seeded annotations have review
  actions, core ABCA1 ATP-coupled phospholipid transporter/floppase activity,
  apoA-I/apoE-linked lipidation and nascent HDL assembly, cholesterol efflux,
  reverse cholesterol transport, and plasma-membrane/endosomal localization are
  defined, and just validate human ABCA1 passes cleanly. ApoA-I-triggered
  Cdc42/cAMP signaling, response/stimulus terms, ER/Golgi/perinuclear
  trafficking, platelet dense-granule and protein-secretion outputs were kept as
  non-core; generic binding and protein-cargo transport labels were
  over-annotated or modified to lipid-specific replacements. Action distribution:
  94 ACCEPT, 34 KEEP_AS_NON_CORE, 11 MARK_AS_OVER_ANNOTATED, 6 MODIFY.
• Completed ABI3 cleanup review: the existing Falcon deep-research-backed ABI3
  review was marked COMPLETE, duplicate defense response to tumor cell
  actions were aligned as non-core, and the Falcon deep-research report was
  cited in the core WAVE regulatory complex synthesis. All 47 seeded annotations
  have review actions, core ABI3/NESH signaling-adaptor membership in the SCAR/WAVE
  regulatory complex, actin-dependent migration/protrusion control, and
  lamellipodium regulation are defined, and just validate human ABI3 passes
  cleanly. Action distribution: 25 ACCEPT, 12 KEEP_AS_NON_CORE, 9
  MARK_AS_OVER_ANNOTATED, 1 UNDECIDED.

2026-06-20

• Completed APP second-pass review cleanup: revised the standalone biological
  description, added local UniProt and Falcon deep-research references, added
  NEW review entries for APP cell-adhesion mediator activity, cell-adhesion
  molecule binding, and copper ion binding, and aligned duplicate-term actions
  for broad protein-metabolism, interleukin-1-response, and amyloid-beta LTP
  annotations. Knowledge-gap questions and suggested experiments now focus on
  evolved in-vivo APP isoform, fragment, adhesion, trophic-signaling,
  metal-binding, and intracellular-domain biology. just validate human APP and
  just render human APP pass cleanly. Action distribution: 251 ACCEPT, 80
  KEEP_AS_NON_CORE, 91 MARK_AS_OVER_ANNOTATED, 3 NEW, 1 REMOVE, 3 UNDECIDED
  across 429 annotations.
• Completed CLU second-pass review cleanup: added
  genes/human/CLU/CLU-notes.md, recorded manual reference-review metadata for
  the Falcon deep-research report, and added knowledge-gap questions plus
  suggested experiments focused on evolved in-vivo CLU extracellular holdase,
  complement MAC-inhibition, receptor-mediated client clearance, glycoform/redox
  regulation, and isoform biology. just validate human CLU and just render human CLU pass cleanly. Action distribution: 73 ACCEPT, 56 KEEP_AS_NON_CORE,
  22 MARK_AS_OVER_ANNOTATED, 4 MODIFY across 155 annotations.
• Completed ABI3 notes backfill: added genes/human/ABI3/ABI3-notes.md to
  capture WRC/adaptor, microglial actin-surveillance, phosphorylation, S209F,
  and model-interpretation evidence from UniProt, primary review citations, and
  the Falcon report. No YAML action changes were needed in this pass; just validate human ABI3 passes cleanly.
• Completed PSEN1 second-pass audit: added reference_review metadata for the
  core gamma-secretase complex/substrate paper (PMID:15274632), the human
  gamma-secretase structure paper (PMID:26280335), and the ER calcium-leak paper
  (PMID:16959576), and appended notes explaining why unusual location
  annotations remain UNDECIDED pending fuller evidence. No annotation action
  changes were made; just validate human PSEN1 and just render human PSEN1
  pass cleanly. Action distribution: 65 ACCEPT, 52 KEEP_AS_NON_CORE, 57
  MARK_AS_OVER_ANNOTATED, 7 MODIFY, 16 UNDECIDED across 197 annotations.
• Completed PSEN2 second-pass audit: added reference_review metadata for APP
  and Notch processing, PSEN2 catalytic aspartate, AP-1-dependent
  late-endosome/lysosome localization, and ER-mitochondria calcium-coupling
  evidence (PMIDs 15274632, 10497236, 10652302, 27293189, 21285369). The
  remaining UNDECIDED calls were left unchanged because they are hypoxia or
  specific synaptic-location transfers without direct cached support. just validate human PSEN2 and just render human PSEN2 pass cleanly. Action
  distribution: 40 ACCEPT, 24 KEEP_AS_NON_CORE, 11 MARK_AS_OVER_ANNOTATED, 4
  UNDECIDED across 79 annotations.
• Completed SORL1 second-pass audit: added reference_review metadata for the
  core APP/SORLA trafficking and amyloid-beta lysosomal-sorting evidence
  (PMIDs 16174740, 16407538, 17855360, 22621900, 24523320), and appended notes
  explaining why the remaining nuclear-envelope style localization annotations
  remain UNDECIDED. No annotation actions were changed; just validate human SORL1 and just render human SORL1 pass cleanly. Action distribution: 92
  ACCEPT, 44 KEEP_AS_NON_CORE, 25 MARK_AS_OVER_ANNOTATED, 4 MODIFY, 3
  UNDECIDED across 168 annotations.
• Completed ADAM10 second-pass audit: added reference_review metadata for the
  main ectodomain-shedding, APP alpha-secretase, synaptic endocytosis,
  L1-cleavage, and S. aureus alpha-toxin context papers (PMIDs 26686862,
  33731436, 23676497, 12475894, 30463011). The existing REMOVE for
  metallodipeptidase activity was retained because ADAM10 is a membrane
  metalloendopeptidase/sheddase, not a dipeptidase. just validate human ADAM10 and just render human ADAM10 pass cleanly. Action distribution: 88
  ACCEPT, 42 KEEP_AS_NON_CORE, 19 MARK_AS_OVER_ANNOTATED, 1 MODIFY, 1 REMOVE
  across 151 annotations.
• Completed INPP5D second-pass audit: reviewed the remaining REMOVE and
  UNDECIDED calls against cached evidence and appended notes. The NKG2A
  protein-binding annotation remains REMOVE because the cached PMID:9485206
  abstract explicitly says SHIP is not associated; the two remaining experimental
  protein-binding annotations remain UNDECIDED because abstract-only evidence
  does not expose SHIP1-specific support. No YAML action changes were needed;
  just validate human INPP5D passes cleanly. Action distribution remains 34
  ACCEPT, 4 KEEP_AS_NON_CORE, 12 MODIFY, 3 MARK_AS_OVER_ANNOTATED, 2 UNDECIDED,
  1 REMOVE, 1 NEW across 57 annotations.
• Completed PLCG2 second-pass audit: reviewed the remaining UNDECIDED and REMOVE
  calls against cached evidence and appended notes. The protein tyrosine kinase
  binding IPI remains UNDECIDED because PMID:32514138 is abstract-only in cache,
  while the Wnt signaling annotation remains REMOVE because PMID:18784435 does
  not support PLCG2 function in Wnt signaling. No YAML action changes were
  needed; just validate human PLCG2 passes cleanly. Action distribution
  remains 83 ACCEPT, 56 KEEP_AS_NON_CORE, 19 MODIFY, 7
  MARK_AS_OVER_ANNOTATED, 1 UNDECIDED, 1 REMOVE across 167 annotations.
• Completed CD33 second-pass audit: reviewed the remaining UNDECIDED
  tau/protein-secretion annotation and appended notes. The PMID:27044754 cache
  names FRMD4A as the functional tau-release hit and does not expose the
  CD33-specific IMP evidence, so the annotation remains UNDECIDED rather than
  removed. No YAML action changes were needed; just validate human CD33 passes
  cleanly. Action distribution remains 29 ACCEPT, 5 KEEP_AS_NON_CORE, 8
  MARK_AS_OVER_ANNOTATED, 2 MODIFY, 1 UNDECIDED across 45 annotations.
• Completed CD2AP second-pass audit: reviewed the remaining UNDECIDED
  tau/protein-secretion annotation and appended notes. The PMID:27044754 cache
  is abstract-only and foregrounds FRMD4A rather than CD2AP-specific evidence,
  so the experimental annotation remains UNDECIDED pending full-text or
  supplementary-data review. No YAML action changes were needed; just validate human CD2AP passes cleanly. Action distribution remains 36 ACCEPT, 9
  KEEP_AS_NON_CORE, 19 MARK_AS_OVER_ANNOTATED, 7 MODIFY, 1 UNDECIDED across 72
  annotations.
• Completed SPI1 second-pass audit: reviewed the two remaining generic
  protein binding IPI annotations against the cached abstracts and appended
  notes. PMID:21575865 foregrounds TMPRSS2:ERG interaction evidence and
  PMID:10207087 foregrounds MEF/AML1 interaction evidence, so both SPI1
  experimental annotations remain UNDECIDED pending full-text or
  supplementary-data review rather than being removed. No YAML action changes
  were needed. Action distribution remains 74 ACCEPT, 32 KEEP_AS_NON_CORE, 21
  MODIFY, 11 MARK_AS_OVER_ANNOTATED, 2 UNDECIDED across 140 annotations.
• Completed NCSTN second-pass audit: added reference_review metadata for the
  core nicastrin/gamma-secretase complex, APP/Notch proteolysis, and structural
  references. No annotation action changes were needed; just validate human NCSTN and just render human NCSTN pass cleanly. Action distribution remains
  73 ACCEPT, 12 KEEP_AS_NON_CORE, 10 MARK_AS_OVER_ANNOTATED, 4 MODIFY across 99
  annotations.
• Completed APH1A second-pass audit: added reference_review metadata for the
  main APH-1/gamma-secretase association, substrate processing, purified-complex
  and structural references. No annotation action changes were needed; just validate human APH1A and just render human APH1A pass cleanly. Action
  distribution remains 50 ACCEPT, 12 KEEP_AS_NON_CORE, 9
  MARK_AS_OVER_ANNOTATED, 2 MODIFY across 73 annotations.
• Completed APH1B second-pass audit: added reference_review metadata for
  APH1B-relevant APH-1 and isoform-specific gamma-secretase references. No
  annotation action changes were needed; just validate human APH1B and just render human APH1B pass cleanly. Action distribution remains 37 ACCEPT, 1
  KEEP_AS_NON_CORE, 4 MARK_AS_OVER_ANNOTATED, 1 MODIFY across 43 annotations.
• Completed PSENEN second-pass audit: added reference_review metadata for
  PEN-2 complex membership, presenilin endoproteolysis, gamma-secretase
  activity, and structural references. No annotation action changes were needed;
  just validate human PSENEN and just render human PSENEN pass cleanly.
  Action distribution remains 55 ACCEPT, 8 KEEP_AS_NON_CORE, 7
  MARK_AS_OVER_ANNOTATED, 1 MODIFY across 71 annotations.
• Completed APOE second-pass audit: added reference_review metadata for the
  key lipid-efflux, astrocyte lipoprotein-release, receptor-mediated
  lipoprotein uptake, amyloid-beta binding, and TREM2-ligand references. No
  annotation action changes were needed; just validate human APOE and just render human APOE pass cleanly. Action distribution remains 145 ACCEPT, 120
  KEEP_AS_NON_CORE, 28 MARK_AS_OVER_ANNOTATED across 293 annotations.
• Completed TREM2 second-pass audit: added reference_review metadata for the
  DAP12 receptor-signaling, apolipoprotein/amyloid/apoptotic-cell ligand,
  myelin-cholesterol response, and human microglia loss-of-function references.
  No annotation action changes were needed; just validate human TREM2 and
  just render human TREM2 pass cleanly. Action distribution remains 117
  ACCEPT, 151 KEEP_AS_NON_CORE, 5 MARK_AS_OVER_ANNOTATED across 273
  annotations.
• Completed ABCA7 second-pass audit: added reference_review metadata for
  phospholipid efflux, apolipoprotein binding, reconstituted phospholipid
  transport, lysophosphatidylcholine export, and APP/amyloid-processing
  references. No annotation action changes were needed; just validate human ABCA7 and just render human ABCA7 pass cleanly. Action distribution
  remains 55 ACCEPT, 20 KEEP_AS_NON_CORE, 4 MODIFY, 2 MARK_AS_OVER_ANNOTATED
  across 81 annotations.
• Completed BACE1 second-pass audit: added reference_review metadata for
  beta-secretase discovery, APP beta-site cleavage, late-Golgi/APP-access
  localization, and GGA/SNX6 trafficking-control references. No annotation
  action changes were needed; just validate human BACE1 and just render human BACE1 pass cleanly. Action distribution remains 94 ACCEPT, 24
  KEEP_AS_NON_CORE, 20 MARK_AS_OVER_ANNOTATED, 2 MODIFY across 140 annotations.
• Completed BIN1 second-pass audit: added reference_review metadata for
  BAR/N-BAR membrane curvature, membrane tubulation, DNM2 interaction,
  BIN1-BACE1 trafficking, and BIN1/tau/clusterin references. No annotation
  action changes were needed; just validate human BIN1 and just render human BIN1 pass cleanly. Action distribution remains 62 ACCEPT, 31
  KEEP_AS_NON_CORE, 18 MARK_AS_OVER_ANNOTATED, 1 MODIFY across 112 annotations.
• Completed CDK5 second-pass audit: added reference_review metadata for p35
  activation, p25 structural context, SH3GLB1/endophilin-B1 autophagy
  phosphorylation, NR2B regulation, and the p35-specific microtubule-binding
  caveat. No annotation action changes were needed; just validate human CDK5
  and just render human CDK5 pass cleanly. Action distribution remains 76
  ACCEPT, 30 KEEP_AS_NON_CORE, 38 MARK_AS_OVER_ANNOTATED, 5 MODIFY across 149
  annotations.
• Completed CDK5R1 second-pass audit: added reference_review metadata for
  p35/CDK5 activation, p35-to-p25 conversion, CDK5/p25 structure, p35
  microtubule binding, p35 myristoylation/localization, and CDK5-mediated
  induced-autophagy context. No annotation action changes were needed; just validate human CDK5R1 and just render human CDK5R1 pass cleanly. Action
  distribution remains 61 ACCEPT, 16 KEEP_AS_NON_CORE, 19
  MARK_AS_OVER_ANNOTATED, 8 MODIFY across 104 annotations.
• Completed notes-only second-pass confirmations for PICALM, CR1, MS4A4A,
  MS4A6A, EPHA1, FERMT2, CASS4, MAPT, GSK3B, LRP1, and ABCA1. These reviews
  already had manual reference_review coverage, no UNDECIDED calls, and no
  REMOVE calls; no YAML action changes were needed. just validate passes
  cleanly for all eleven genes. Action distributions remain: PICALM 80 ACCEPT,
  22 KEEP_AS_NON_CORE, 10 MARK_AS_OVER_ANNOTATED, 1 MODIFY; CR1 42 ACCEPT, 17
  KEEP_AS_NON_CORE, 2 MODIFY, 3 MARK_AS_OVER_ANNOTATED; MS4A4A 7 ACCEPT, 2
  MARK_AS_OVER_ANNOTATED; MS4A6A 3 ACCEPT, 1 MODIFY, 2
  MARK_AS_OVER_ANNOTATED; EPHA1 30 ACCEPT, 10 KEEP_AS_NON_CORE, 2
  MARK_AS_OVER_ANNOTATED; FERMT2 52 ACCEPT, 30 KEEP_AS_NON_CORE, 3 MODIFY; CASS4
  13 ACCEPT, 1 MODIFY, 1 KEEP_AS_NON_CORE; MAPT 77 ACCEPT, 70 KEEP_AS_NON_CORE,
  50 MARK_AS_OVER_ANNOTATED, 1 MODIFY; GSK3B 172 ACCEPT, 126
  KEEP_AS_NON_CORE, 84 MARK_AS_OVER_ANNOTATED, 8 MODIFY; LRP1 77 ACCEPT, 29
  KEEP_AS_NON_CORE, 18 MARK_AS_OVER_ANNOTATED; ABCA1 94 ACCEPT, 34
  KEEP_AS_NON_CORE, 11 MARK_AS_OVER_ANNOTATED, 6 MODIFY.
• Created and validated modules/alzheimer_disease_pathways.yaml, a
  ModuleReview overview module grouping the reviewed gene set into APP
  processing/amyloid-beta handling, lipid/lipoprotein transport, microglial
  lipid-debris sensing, tau/cytoskeletal kinase biology, and
  endocytic/adhesion adaptor systems. uv run linkml-validate -s src/ai_gene_review/schema/gene_review.yaml -C ModuleReview modules/alzheimer_disease_pathways.yaml passes cleanly, and the module renders
  to pages/modules/alzheimer_disease_pathways.html.