Human Proteostasis Network Project

Background

Proteostasis is used here in the broad systems sense: the cellular machinery that
supports protein synthesis, folding, trafficking, quality control, sequestration,
and degradation.

The source resource analyzed in this project is the Human Proteostasis Network
Annotation 4.3.11 workbook together with three Proteostasis Consortium survey
manuscripts:

• MS1 introduces the overall PN framework and covers translation, folding,
  transport, and organelle-specific proteostasis systems.
• MS2 covers the autophagy-lysosome pathway (ALP) and provides the most
  detailed row-level notes and references in the current release.
• MS3 covers the ubiquitin-proteasome system (UPS) and explains the
  domain-heavy inclusion and classification logic used for that branch.

The mapping and browser artifacts now use the 2026-04-17 PN 4.3.11
workbook. Earlier 2024 release artifacts are kept only as provenance for
individual curation comments that were started against that release.

Overview

The Proteostasis Consortium workbook and papers define a curated human proteostasis
membership and role taxonomy. They do not define a GO-ready annotation set.

This project asks three main questions:

• What biological claims do the PN annotations actually make?
• Which PN statements are close to GO terms, which imply ontology gaps, and which
  are better treated as systems metadata?
• Which workbook entries look especially useful, curious, or suspect in the
  context of AIGR?

Within this project:

• the PN resource provides systems architecture and candidate roles
• unfolded protein binding is treated as one mechanistic subdomain within proteostasis
• AIGR uses PN as a scaffold, a QA source, and a prioritization layer

See:

• Priority genes
• PN tree browser
• ALP mapping set
• Chaperone mapping set
• ER proteostasis mapping set
• Extracellular proteostasis mapping set
• Mitochondrial proteostasis mapping set
• Nuclear proteostasis mapping set
• PN regulation mapping set
• Translation mapping set
• UPS mapping set
• Project-local tests
• Project-local reports
• Report-local PN tree browser
• Mapping scrutiny report
• Unusual propagation report
• Mapping export workbook
• UPB gene list
• Unfolded Protein Binding project
• Ribosome Quality Control project
• Integrated Stress Response project
• ER-phagy project

What The PN Resource Actually Contains

The workbook is a row-per-role annotation table, not a gene-centered review file.

• 3123 unique genes
• 4000 annotation rows
• 323 genes with cross-branch annotations
• 318 genes with multiple annotations within one branch
• 2482 genes with a single annotation

The custom hierarchy has:

• 9 Branches
• 33 Classes
• 277 Groups
• 676 Types
• 536 Subtypes

This is their own vocabulary:

• the workbook contains no GO IDs or explicit GO term mappings
• the papers explicitly describe the PN hierarchy as a taxonomy that complements structured vocabularies such as GO
• the manuscripts say GO, Reactome, KEGG, UniProt, and InterPro were used as inputs for building preliminary lists, not as the target representation

Current Mapping Completion Status

The current curated mapping pass is complete for the Human Proteostasis Network Annotation 4.3.11 workbook release dated 2026-04-17.

Coverage after the completion pass:

Level	Total source codes	Pending review	Mapped	Context only	No mapping	Deferred	Missing from YAML
Branch	9	8	0	1	0	0	0
Class	42	30	9	2	1	0	0
Group	297	229	60	2	1	5	0
Type	800	661	119	2	14	4	0
Subtype	881	795	64	1	14	7	0

Every 2026-04-17 PN source code now has exactly one subject_curations
record in a branch mapping YAML. missing_from_yaml is now a QA failure state,
not a normal curator bucket, and should remain zero.

The current hierarchy has 2029 total source nodes and 1348 leaf nodes.
The YAML inventory contains:

Curation status	Records	Meaning
pending_review	1723	Accounted for in YAML, but not yet manually analyzed in depth
mapped	252	Reviewed and mapped to a GO term
context_only	8	GO relationship recorded, but unsafe for gene-level propagation
no_mapping	30	Reviewed and concluded that no GO mapping should be made
deferred	16	Reviewed, but blocked by evidence, taxonomy ambiguity, or a missing/better GO term

Mapping scopes are:

Mapping scope	Records	Use
exact	4	Direct semantic match
ok_for_propagation_to_go	248	May produce candidate gene-GO propagations
too_broad_to_propagate	8	Real contextual alignment, but excluded from propagation

Most PN source codes are not yet final non-map calls. They are explicitly
tracked as pending review so curators can distinguish coverage bookkeeping from
completed curation decisions.

Projection against the human GOA DuckDB at
~/repos/go-db/db/goa_human.ddb produced:

Projection status	Unique gene-GO pairs
already in GOA exactly	703
entailed by GOA closure	403
more specific than existing GOA	238
supported by GOA regulation	77
new to GOA	760
no local GOA available	16

Only the 1075 candidate additions (more_specific_than_existing_goa +
supported_by_goa_regulation + new_to_goa) should enter manual AIGR
rereview queues. The no_local_goa class is mostly a data-availability state,
not biological evidence; with the DuckDB source it is now a small residual
category.

Extra-Scrutiny Findings

The mapping audit flags 180/260 GO-bearing curation records as requiring manual
gene-level review before they are used to change a gene review. These are not
necessarily wrong mappings; they are places where propagation can mislead if the
projected GO term is treated as an asserted gene function.

Main flagged patterns:

• 151 mappings have regulatory, recruitment, localization, sensing, or other
  contextual PN source labels.
• 73 mappings use broad or context-losing GO targets such as generic
  translation, protein transport, DNA repair, DNA binding, or stress-response
  terms.
• 44 mappings include domain, family, or subtype metadata in the source label.
• 12 mappings are at branch or class level.
• 8 mappings are explicitly categorized as too_broad_to_propagate and are
  excluded from propagation reports.

Representative cases that should stay in the manual review queue:

• Translation -> GO:0006412 translation: useful as high-level context but too
  broad for many PN rows; [EDF1](../../genes/human/EDF1/EDF1-ai-review.html) showed that only the RQC term survived manual
  rereview.
• Mitochondrial proteostasis|Protein transport|Protein import -> GO:0017038 protein import: can be broader than an existing route-specific
  mitochondrial import term, as seen for [TOMM20](../../genes/human/TOMM20/TOMM20-ai-review.html).
• Autophagosome/endosome docking and related ALP docking labels ->
  GO:0061909 autophagosome-lysosome fusion: plausible pathway-stage
  propagation, but [RAB7A](../../genes/human/RAB7A/RAB7A-ai-review.html) showed that fusion versus post-fusion maturation must
  be checked gene by gene.
• HSPA8-like CMA/aggrephagy boundary cases: PN aggregate-handling context does
  not automatically justify aggrephagy when direct CMA annotations are better
  supported.
• UPS ubiquitin/UBL-binding context buckets: many are useful triage labels, but
  the UPS branch is intentionally inclusive and domain-heavy, so these should
  not be imported into GO without independent evidence.

Evidence Shape By Branch

The current release is uneven by design.

Branch set	What is present in the workbook	What it means for reuse
ALP	Per-row notes for 1003/1003 rows and references for 1001/1003 rows	Best-supported branch for row-level reuse and audit
UPS	Principal domains for 1528/1528 rows and auxiliary domains for 1521/1528 rows	Strong domain/family scaffold, but weaker row-level functional justification in this release
Other 7 branches	No row-level notes, no row-level references, no domain columns	Useful as curated membership/context, but not ready to import into GO as-is

What Each Level Means

The hierarchy is semantically mixed, which is the main reason it does not map cleanly to GO.

PN level	What it usually encodes	Relation to GO
Branch	Localization or top-level pathway membership	BP/CC hybrid, not a GO class
Class	Function in proteostasis, except ALP where it is a stage of autophagy	MF/BP hybrid
Group	System, complex, pathway module, or mechanistic bucket	Sometimes GO-like, often not
Type	More specific mechanistic role, family, or complex membership	May correspond to MF, BP, CC, or family metadata
Subtype	Structural family, domain class, or finer mechanistic subdivision	Often outside GO scope

Examples by level:

• Branch
• Cytonuclear proteostasis
• ER proteostasis
• Autophagy-Lysosome Pathway
• Ubiquitin Proteasome System
• Class
• Chaperone
• Protein transport
• Autophagophore initiation and elongation
• E3 ubiquitin and UBL ligases
• Group
• HSP70 system
• TRAP complex component
• Class 3 PI3K complex 1, direct
• CRL family
• Type
• HSP70 nucleotide exchange factor
• J-domain containing HSP70 cochaperone
• Modulator of class 3 PI3K complex 1 activity
• BTB
• Subtype
• BAG domain family
• KCTD type
• WD40
• many rows have no subtype

Fully expanded examples:

• [BAG1](../../genes/human/BAG1/BAG1-ai-review.html)
• Cytonuclear proteostasis -> Chaperone -> HSP70 system -> HSP70 nucleotide exchange factor -> BAG domain family
• [STAT1](../../genes/human/STAT1/STAT1-ai-review.html)
• Autophagy-Lysosome Pathway -> Autophagy gene expression -> Transcriptional repressor
• [KCTD11](../../genes/human/KCTD11/KCTD11-ai-review.html)
• Ubiquitin Proteasome System -> E3 ubiquitin and UBL ligases -> CRL family -> BTB -> KCTD type

These are informative for humans, but they are not all GO terms and they are not all the same kind of thing.

Relationship To GO

Explicit GO mapping

There is essentially none in the released workbook.

• I searched the workbook for GO: and Gene Ontology strings and found none.
• The row schema is PN-native: Branch/Class/Group/Type/Subtype, plus optional notes, references, and UPS domain fields.
• The papers cite GO as a source database used to build candidate lists, especially for ALP, not as a field embedded back into the final annotation table.

Implicit GO mapping

There is a lot of implicit overlap with GO, but it is inconsistent in kind.

Some PN rows look close to GO:

• TRAP complex component
• Class 3 PI3K complex 1 component
• HSP70 nucleotide exchange factor
• Ribosome-associated QC

But many PN labels are not GO-ready assertions:

• family labels like BAG domain family
• domain labels like WD40 or TPR domain containing
• branch-local staging labels like Autophagophore initiation and elongation
• inclusive UPS buckets like RING, BTB, DCAF, MEX3

So the right reading is:

• the PN taxonomy is biologically meaningful
• some rows are close to GO MF/BP/CC concepts
• many rows are module, family, or pathway-context labels rather than GO-annotatable facts

Strategy For Bringing PN Into GO

This section is independent of AIGR workflow.

1. Separate membership from GO assertion

A PN row first tells us:

• this gene belongs in proteostasis
• the authors place it in a particular proteostasis module

It does not automatically tell us:

• the correct GO aspect
• the exact GO term
• whether the role is core vs contextual
• whether the evidence is strong enough for manual GO curation

2. Tag each row by provenance

At minimum, rows should be split into:

• entity_based_literature_supported
• domain_based_family_inference
• ALP_note_backed
• UPS_domain_backed
• cross_branch_context_only

This is the biggest determinant of whether a row is GO-ready, prediction-like, or just useful context.

3. Decompose each PN row into candidate GO semantics

Each row should be classified into one of four buckets:

1. exact_or_near_GO
   Example: complex component, transport process, clear enzymatic role.
2. GO_with_context_loss
   Example: a GO term exists but loses the proteostasis-system framing.
3. ontology_gap
   Example: the biology is real but current GO lacks a clean term or term family.
4. non_GO_metadata
   Example: family/domain/subtype labels that should stay as supporting metadata.

4. Curate by branch, not all at once

The branches are not equally reusable.

• Start with ALP rows that have notes and references.
• Use non-ALP/non-UPS branches to identify likely GO-compatible roles and obvious positive controls.
• Treat UPS as a mixed case: many rows are probably best handled as prediction candidates or review leads until branch-specific evidence is inspected.

5. Keep domain-based rows conservative

The papers explicitly flag domain-based inclusions and borderline cases.

Concrete examples:

• [HSPA12A](../../genes/human/HSPA12A/HSPA12A-ai-review.html) and [HSPA12B](../../genes/human/HSPA12B/HSPA12B-ai-review.html) are included as HSP70-family PN components even though MS1 says their proteostasis functions are not yet known.
• UPS authors explicitly say they aimed to be inclusive and that domain weight is debatable in some cases.

These should not be imported into GO as manual curation without independent evidence.

6. Expect ontology-gap work

The PN resource reinforces that GO has missing or awkward coverage around:

• holdase chaperoning
• sensor/adaptor roles in quality control
• co-chaperone mechanistic MF space
• proteostasis-system context that mixes process and molecular role

AIGR Triage Opportunities

The PN resource is useful inside AIGR primarily as a triage and QA layer.

High-value positive controls

These are cases where PN and existing AIGR work largely agree and help validate the framework:

• [CRYAA](../../genes/human/CRYAA/CRYAA-ai-review.html), [CRYAB](../../genes/human/CRYAB/CRYAB-ai-review.html): PN small-HSP placement agrees with the UPB/holdase interpretation
• [SSR1](../../genes/human/SSR1/SSR1-ai-review.html), [SSR2](../../genes/human/SSR2/SSR2-ai-review.html): PN TRAP complex component fits existing ER protein-targeting/translocation GO work
• [EDF1](../../genes/human/EDF1/EDF1-ai-review.html): PN ribosome-associated QC placement is consistent with the current AIGR review
• [KCTD11](../../genes/human/KCTD11/KCTD11-ai-review.html): PN CRL/BTB placement matches the current AIGR view of a Cul3 adaptor

Context-specific but plausible PN rows

These are probably real, but should be treated as non-core or secondary until checked carefully:

• [STAT1](../../genes/human/STAT1/STAT1-ai-review.html): ALP transcriptional repressor role based on ULK1 promoter regulation
• [BIRC6](../../genes/human/BIRC6/BIRC6-ai-review.html): ALP docking/fusion regulator role on top of its core UPS E2/E3 biology
• [CIAO1](../../genes/human/CIAO1/CIAO1-ai-review.html): UPS DCAF placement on top of its stronger Fe-S assembly/chaperone identity

Domain- or family-driven caution cases

These are especially useful for AIGR because the papers themselves signal uncertainty or inclusivity:

• [HSPA12A](../../genes/human/HSPA12A/HSPA12A-ai-review.html), [HSPA12B](../../genes/human/HSPA12B/HSPA12B-ai-review.html): included as HSP70 PN components despite admitted lack of clear proteostasis function
• [AARSD1](../../genes/human/AARSD1/AARSD1-ai-review.html): dual placement as HSP90 cochaperone and tRNA synthetase
• [BAG6](../../genes/human/BAG6/BAG6-ai-review.html): split across GET-pathway transport and UPS; MS1 explicitly distinguishes it from canonical BAG-domain NEFs
• [TTC28](../../genes/human/TTC28/TTC28-ai-review.html): PN places it as an HSP70-HSP90 joint cochaperone, but the current AIGR review treats it as a mitotic scaffold
• [MEX3B](../../genes/human/MEX3B/MEX3B-ai-review.html): RING-family UPS placement may be real but needs distinction between family membership and core proteostasis role

Existing-review rereview examples

On the 83-gene existing-review queue, the useful distinction was not just the
pipeline label but whether the PN-projected term was actually a better GO
assertion than the current AIGR review. In practice,
more_specific_than_existing_goa is a projection label, not a guarantee that
the PN term remains the most specific biologically defensible choice after
manual rereview.

• [BCAP31](../../genes/human/BCAP31/BCAP31-ai-review.html) (more_specific_than_existing_goa): accepted GO:0036503 ERAD pathway.
  This is a good positive-control case where the PN mapping is exact and the
  biology holds up. The current review already supported direct ERAD
  participation, and the rereview added the pathway term because BCAP31 helps
  handle retrotranslocation of ERAD substrates rather than only acting in a
  looser ER-stress or regulatory context.
• [EDF1](../../genes/human/EDF1/EDF1-ai-review.html) (new_to_goa): PN suggested GO:0002181 cytoplasmic translation,
  GO:0006412 translation, and GO:0006515 protein quality control for misfolded or incompletely synthesized proteins. Only GO:0006515 survived
  conservative rereview. The broad translation terms were not added because the
  best-supported biology is collided-ribosome surveillance and
  ribosome-associated quality control rather than generic translation.
• [TOMM20](../../genes/human/TOMM20/TOMM20-ai-review.html) (more_specific_than_existing_goa in the queue): rejected. The PN
  mitochondrial mapping propagates the group-level Protein import bucket to
  GO:0017038 protein import, but the current AIGR review already uses the
  route-specific mitochondrial import term GO:0030150 protein import into mitochondrial matrix. At the gene-review level, the PN suggestion was
  broader rather than more specific.
• [HSPA8](../../genes/human/HSPA8/HSPA8-ai-review.html) (more_specific_than_existing_goa in the queue): rejected. The PN
  GO:0035973 aggrephagy projection comes from a selective-autophagy-receptor
  path, whereas the current review already captures HSPA8's direct and much
  better supported CMA biology with GO:0061684 chaperone-mediated autophagy
  and GO:0061740 protein targeting to lysosome involved in chaperone-mediated autophagy. HSPA8 clearly participates in proteostasis and aggregate handling,
  but that did not justify promoting it to aggrephagy here.
• [RAB7A](../../genes/human/RAB7A/RAB7A-ai-review.html) (more_specific_than_existing_goa in the queue): rejected on
  conservative rereview. PN projected GO:0061909 autophagosome-lysosome fusion, but the local evidence base is mixed, and mammalian knockout work
  supports a stronger role in post-fusion autolysosome maturation than in the
  fusion step itself. That was not strong enough to add the more specific term
  to the human review.

Priority Review Targets

See priority_genes.tsv.

Recommended first-pass jobs:

1. [BTF3](../../genes/human/BTF3/BTF3-ai-review.html): fetch and review the true human [BTF3](../../genes/human/BTF3/BTF3-ai-review.html) gene (P20290) as the PN
   nascent-polypeptide-associated complex component.
2. [HSPA12A](../../genes/human/HSPA12A/HSPA12A-ai-review.html) and [HSPA12B](../../genes/human/HSPA12B/HSPA12B-ai-review.html): fetch and review as explicit domain-based PN inclusions.
3. [AARSD1](../../genes/human/AARSD1/AARSD1-ai-review.html): review the dual chaperone/translation placement.
4. [BAG6](../../genes/human/BAG6/BAG6-ai-review.html): review as a multi-branch boundary case connecting transport, UBL biology, and proteostasis.
5. [TTC28](../../genes/human/TTC28/TTC28-ai-review.html): explicitly test the PN cochaperone claim against the existing mitosis-focused review.

PN-vs-UPB Comparison

• all 33/33 human UPB genes are present in the PN workbook
• most of those genes are consistent with PN placement at a coarse level
• the best bridge cases are the small HSPs, HSP70/J-domain systems, and the QC boundary genes

The UPB project is the best place to reason about:

• direct unfolded-protein binding
• holdase vs foldase vs co-chaperone distinctions
• GO MF correction and ontology-gap pressure

The PN project is broader:

• define the proteostasis universe
• understand the authors' own annotation model
• identify GO-compatible vs non-GO-compatible PN statements
• use PN to drive AIGR QA and job selection

Next Steps

• Audit the highest-priority genes in priority_genes.tsv.
• Work through the 1075 projected candidate additions, using the unusual
  propagation report as a blocklist for automatic review edits.
• Promote only gene-level decisions that survive evidence review into AIGR
  YAML; leave broad PN context as project metadata.