Human ALB is HGNC:399 / UniProt P02768. All 66 machine-seeded annotations, 58 original reference identities and three alternative-product records were preserved. Every row was manually assessed. There are 36 ACCEPT, nine KEEP_AS_NON_CORE, four REMOVE, three MARK_AS_OVER_ANNOTATED, one MODIFY and 13 UNDECIDED decisions; no NEW rows. The two integrated cores are extracellular molecular carriage and antioxidant protection. Every core term was already present in the seed; this review creates no new process assertion.
The seed is the genuine source4 runner output, not an authored substitute. Its canonical import and subsequent auxiliary import were reviewed separately. The latter admitted exactly 17 missing publication records and 17 missing Reactome records, with exclusive creation and byte/hash verification. Sixteen existing publication records, four existing Reactome records and InterPro files were preserved; in particular, divergent newer-looking records were not substituted for existing caches. Main absence was checked at fb53b36b6141d82016b012f1d45c6d1dc9901541; the publication base is the imported 39b086a55856ca2f2287036a5ab43280add1ee90 snapshot. The coordinator rechecks main before publication.
The approved HGNC name has no listed previous symbols or aliases. Canonical and twelve historical UniProt-name paths were checked independently, with no conflicting directory or open title match. Raw UniProt, GOA and alternative products remain unchanged. Their bibliography is not automatically promoted to reviewed evidence: only sources actually used below or in the YAML enter this review's citation census.
The default Falcon research command with --fallback perplexity-lite --timeout 1200 was actually attempted using ordinary writable UV tool/cache directories. An offline launch could not resolve the uncached client; the normal online attempt then failed for both providers during dependency resolution because pypi.org DNS failed. No provider report was produced or fabricated. Manual primary-source research continued. The ordinary publication-cache command subsequently completed successfully, reporting 33/33 original gene publications already cached. The added mouse-donor source was already cached. Research logs, source receipts and immutable-baseline hashes are retained in the publication manifest's evidence paths.
Albumin's ligand pockets support soluble carriage rather than transmembrane pumping. Bilirubin equilibrium binding was measured with purified human albumin by ultrafiltration PMID:11397817 and with normal human serum by charcoal titration PMID:1157293. Affinity and capacity depend on conditions; neither experiment supplies one universal physiological dissociation constant. Human albumin/myristate structures PMID:9731778, oleate NMR/displacement PMID:16289007 and myristate-associated conformational studies PMID:16413837 give direct fatty-acid evidence. Seventeen crystallographic human-albumin drug/toxin complexes support toxin binding PMID:16169013. These specific activities fit the integrated carrier core without manufacturing redundant NEW rows.
The original human-protein study PMID:20705604 positively resolves an extracellular hemin acceptor. Original indexed PMC2963377, Methods and Results/Figure 8, were recovered with searches for "PMC2963377" "Hemin Transfer" "HSA" and "PMC2963377" "albumin" "Methods"; direct opening was intermittently challenged. Recombinant human ABCG2 ECL3/full transporter and albumin were examined. Apoalbumin reversed ECL3 fluorescence quenching, hemin-loaded albumin did not, and a Soret shift supported ligand transfer. Albumin accepts completed heme; ABCG2 is the transporter. This is stronger than merely inferring albumin function from the paper's ABCG2-focused title. The normal cache remains abstract-only.
Human albumin and the DAHK N-terminal peptide protected murine cortical cultures against hydrogen peroxide and copper/ascorbate PMID:14726550. The protein species is human and the host culture is mouse. The original abstract gives a concentration-dependent protective result, while attributing it probably to metal binding. Antioxidant activity is retained without claiming every potential thiol, heme or peroxide catalytic mechanism was directly assayed. Live GO relates antioxidant activity to cellular oxidant detoxification; the process does not require the participating protein to be inside the cytosol.
The 2023 full narrative review PMID:36979511 was read as a synthesis, not as a new experimental assay despite the immutable seeded EXP code. Its heme-binding conclusion has independent original support above. One bounded background error is explicitly excluded: section 1 says heme is part of cyanocobalamin. ChEBI's cyanocob(III)alamin definition describes a cobalt-centered corrin, not iron heme. This error is not generalized into a rejection of albumin heme binding, and unrelated speculative review claims are not adopted.
Rice starvation: the full original author-uploaded paper for PMID:16245148 was read, including Methods and Figures 1, 3–6. Mature human ALB coding sequence from human blood RNA was placed behind the rice alphaAmy3 promoter/signal sequence and untranslated regulatory sequence. Starved transformed rice cells produce and secrete HSA. The rice regulatory system performs the response; ALB is its expressed output. The human starvation-response annotation is removed on this verified experimental scope, not because the title names rice or because albumin expression can never respond to nutrients. An independent annotation-reviewer consultation recovered the same original Methods/Results and agreed with this distinction.
Engineered oxygen carrier and tetramer: the positive abstract PMID:16283771 explicitly describes chemical crosslinking of recombinant human albumin through lysine/Cys34 groups and loading with synthetic Fe(II) porphyrins. These artificial assemblies reversibly bind oxygen. The oxygen row's contributes_to flag is preserved: albumin contributes the designed scaffold, but this still does not establish oxygen binding by native albumin. Both this native-function inference and the endogenous protein-complex inference are marked over-annotated. The whole Methods were not recovered, and no claim is made that the reported artificial-material results are false. Independent consultation confirmed this bounded interpretation of the abstract.
Heme pathways: Reactome heme biosynthesis includes delivery of completed heme. The albumin handoff does not synthesize its tetrapyrrole, so that process assertion is over-annotated. Heme degradation distinguishes heme-oxygenase cleavage and biliverdin reduction from the downstream bilirubin handling performed by albumin. That source is refined to bilirubin transport, supported by BIL binds ALB. Heme transfer to hemopexin R-HSA-2168887 and binding/release of bilirubin, lipids and drugs remain positive carrier evidence. This does not prohibit scaffolds from process participation; it identifies the actual step performed here.
Protein associations: human ovarian-cancer serum albumin-associated BRCA2 fragments were verified in original Results/Figures 4–5 PMID:16099937. The study resolves fragments rather than full-length BRCA2 and does not demonstrate delivery to a particular acceptor. The generic protein-binding row is removed as uninformative, without denying association or inventing protein-carrier activity. The human plasma albumin–alpha1-microglobulin complexes PMID:9183005 are likewise positive interactions without a demonstrated new specific MF. The abstract of PMID:23384347 does not expose its albumin pair; no wrong-gene inference is made. Its generic term is removed for lack of functional information, not as evidence that the interaction is false.
VvhA stabilization: the complete three-page J-STAGE original PDF PMID:16394536 was inspected. Human/bovine albumin was added to Vibrio culture supernatant; human erythrocyte lysis and VvhA monomer/oligomer immunoblots were measured. Native PAGE and unheated SDS-PAGE Figures 2–3 support stabilization and altered oligomerization. The inspected assays do not resolve an albumin–VvhA bound pair, leaving the exact exogenous-protein-binding MF UNDECIDED. No holdase or host-defense MF is inferred. The public source is documented here; supporting_text_fulltext is not used as a workaround for an uncached public article.
Extracellular region is retained across the direct, automatic and Reactome assertions. The positive secreted UniProt record and human serum experiments independently establish this broad core location. Broadness alone is not a source failure. Each Reactome row was assessed against its specific ligand-binding/release or substrate event; the normal cached summaries do not purport to be experimental localization images. Official event pages provide the compartment context.
ER lumen in the FAM20C substrate event and platelet alpha-granule lumen are contextual pools. Albumin does not thereby become a kinase, granule assembly factor or exocytosis motor. The mouse cytoplasm transfer is consistent with the verified MGI ortholog graph and original mouse embryo/teratocarcinoma immunostaining, transcripts and uptake PMID:8607965. Cytoplasm includes the secretory compartment and does not mean cytosol. These pools are NON_CORE.
Purified HSA–calf-thymus-DNA interactions PMID:16405401 are retained as conditional DNA binding without a nuclear regulatory core. Cys34-dependent human-albumin dimerization/aggregation PMID:26554815 is retained as conditional self-binding; the independent serum-interactome source PMID:15174051 remains UNDECIDED because its target-specific self-association controls were not recovered. Human stressed-plasma albumin is a client of clusterin, supported by affinity capture/MS and sandwich ELISA PMID:19996109. This establishes chaperone binding by albumin, not chaperone activity.
Human/bovine albumin binding to enterobactin and ferric enterobactin PMID:6234017 is directly supported by spectroscopy, stoichiometry and bacterial iron-utilization effects, distinct from aerobactin. Both source-specific IPI and mammalian IBA are retained as NON_CORE. The English abstract PMID:16201370 says vitamin B6 but does not resolve its exact chemical form as pyridoxal phosphate; the original Chinese full paper was not recovered, so PLP binding remains UNDECIDED.
The human-cell/albumin-species distinction matters for PMID:16153637. Its abstract describes human SH-SY5Y cells, TMRE protection and unchanged calcium-release kinetics. PubMed indexing and the same authors' original conference abstract P251 identify bovine albumin. The complete journal Methods were not recovered to establish whether human albumin was additionally assayed. Both the human mitochondrial-depolarization and calcium-starvation process rows remain UNDECIDED; the positive protective result is preserved. Host-cell identity alone cannot resolve protein species.
For proteomic nucleus PMID:21630459, plasma microparticles PMID:22516433, and exosomes PMID:19056867, PMID:20458337, PMID:21276792, PMID:21362503, PMID:23533145, the ALB-specific hit/association controls were not independently recovered. The latter two have locally available body text, but target supplements are still unresolved. These rows are UNDECIDED, without treating serum abundance as proof of contamination. Broader extracellular assertions from plasma PMID:14718574, infant nasopharyngeal fluid PMID:21805676 and tears PMID:22664934, PMID:23580065 remain sound from independent positive albumin secretion/serum evidence; acceptance is not a claim that the individual ALB peptide table was inspected. The glaucoma paper's original publisher record independently verifies title, DOI and abstract.
Live AmiGO definitions were read for molecular carrier activity, protein carrier activity, heme biosynthetic process, heme catabolic process, cellular response to calcium starvation, response to nutrient levels, oxygen binding and cellular oxidant detoxification. Carrier function requires carriage to an acceptor or location, not merely an arbitrary protein interaction. Heme catabolism describes ring breakdown. Nutrient-response definitions include expression changes, so the rice correction turns on which gene product performs the response rather than a blanket exclusion of expression evidence. Antioxidant activity has the verified part_of link used by the seeded logical inference.
The cached PTHR11385 PAINT records explicitly identify nutrient-response node PTN000147344 and mammalian enterobactin-binding node PTN008319942. Only the PTN node is entered in each IBA propagation block. The target's appearance among descendants is valid experimental grounding, not circularity. One defective human starvation source does not establish that the ancestral node is wrong; other descendant assays and the full node/MSA placement remain unresolved. The enterobactin node has direct human/bovine grounding. Compara donor identifiers and combined InterPro/ARBA/location mappings are preserved; uninspected individual rule conditions are not relabeled direct evidence.
No P02768 activity was found in the local GO-CAM index. That is an access/inventory result, not evidence of a biological gap. Neither a redundant parent/child nor a new process is proposed. The core carrier process is the existing bilirubin-transport assertion; the antioxidant process is the existing logical oxidant-detoxification assertion. The two cores keep species and compartment boundaries explicit.
The finite authored-source census contains 34 PMIDs (33 original plus the already cached mouse-donor study), 21 Reactome records, four original GO reference methods, the unchanged UniProt record and these notes. All PMID and Reactome cache paths are present. Four publication records expose body text locally; the other 30 are abstract-only. Availability flags describe those machine records. External primary reading does not change local cache flags, and partial body extraction is not described as complete absence of full text. There is no generated provider file or provider bibliography to expand.
Every DOI/PMC URL used in these notes resolves to an already named original PMID. The separate conference PDF is a same-author source-access receipt for protein-species uncertainty, not a new journal citation or an independently seeded functional claim. UniProt's unused bibliographic list, review back-references and publications merely encountered in search are not silently counted as inspected scientific dependencies. All original identifiers/titles remain intact, including the review article's seeded evidence code. Validation results and the exact recursive citation/hash census are recorded in the final manifest.
Targeted full validation completed with exit 0 and no review advisories; the runtime emitted only a dependency deprecation notice. All 92 supporting passages match their actual local source with case-sensitive whitespace normalization, with no match relying on case folding. All 66 source assertions, 58 original reference identities, three alternative products, raw GOA/UniProt files and 34 imported cache byte strings pass preservation checks. The YAML contains no anchors or trailing spaces. History validation and rendering pass. COMPLETE describes the finished manual audit, not certainty about the 13 explicitly UNDECIDED source assertions.
The current formal review 5330691370 and full comment 5856716505 were read.
The heme-biosynthesis action is changed from MARK_AS_OVER_ANNOTATED to
REMOVE, matching its existing source-specific rationale. In PMID:20705604,
human albumin accepts already formed hemin from ABCG2's extracellular loop;
it does not perform a step of heme formation. The positive carrier evidence
is preserved, and no replacement activity or NEW process is invented. This
supersedes the earlier note's over-annotation action for that pathway.
Bilirubin transport remains the supported refinement of the separate heme
catabolism row.
The suggested question on colloid osmotic pressure records a physiological
role already stated in the immutable UniProt FUNCTION section, which labels
it Probable. It asks how concentration and ligand occupancy affect that
contribution; it does not assert an unverified ontology gap or add a core
term. The seven discussed proteomic location rows remain UNDECIDED because
the source-specific albumin evidence was not recovered. Comparisons with
other genes and albumin abundance do not establish contamination or override
that evidence limit. The actual child event R-HSA-9709883 remains valid
support for the broader parent R-HSA-9707564 assertion.
All 66 annotation source objects, two cores, three products and 61 reference
identities are preserved. The updated actions are 36 ACCEPT,
9 KEEP_AS_NON_CORE, 5 REMOVE, 2 MARK_AS_OVER_ANNOTATED, 1 MODIFY and
13 UNDECIDED. No new source or cache edit is required.