An evolving AI research projectBackground · goals · methodsUpdated 11 Oct 2026

Approaches to aging

Cumulative update · 11 Oct 2026 Two GEO series from one paper study old TIMP2-treated wild-type mice and neonatal microglia with different genotypes. They cannot identify a joint genotype-by-treatment effect. AG-T01 remains 🔵 Level 2. Daily report 11 Oct 2026 →

What may counter age-related functional decline · Updated 28 September 2026

This list grows with the evidence. Its first entries concern human function; subsequent entries are mechanistic candidates or untested hypotheses. Inclusion does not establish slowing of human aging. The order is not a personal treatment recommendation.

AG-A01 · Structured physical activity

Human function · Project level 1

Outcome / target: Less major mobility disability than with health education.
Population / model: Older adults at mobility risk.
Evidence: Randomised; functional endpoint.
Limits / counterevidence: Serious harms not excluded; no general aging reversal.
Next desk check: Check adherence, safety and separate long-term replications.
Sources: AG-J01 · 2014-05-27

AG-A02 · Combined programs for cognition

Human function · Project level 1

Outcome / target: Small composite-test advantages over each study’s comparator.
Population / model: People at elevated cognitive risk.
Evidence: Two randomised cohorts; related programs.
Limits / counterevidence: Components unresolved; dementia prevention unproven. Assess burdens.
Next desk check: Check contact time, practice effects, daily relevance and counterstudies.
Sources: AG-J02 · 2015-03-12; AG-J03 · 2025-07-28

AG-A03 · TIMP2 as a functional candidate

Preclinical · Project level 1

Outcome / target: Target: a direct functional benefit versus control with preserved connections. AG-R05 reports a direct cognitive group comparison, with magnitude and precision unresolved here. AG-R08 Y-maze tests against chance alone do not provide that contrast; nesting remains separate.
Population / model: Mouse.
Evidence: Preclinical; mechanism unresolved.
Limits / counterevidence: Direct neuronal action competes with microglial mediation. Also separate degradation from export; lower cellular loading does not establish safe disposal. The new tau source is a countercontext, not TIMP2 evidence. 30 Sep: the audited assays do not identify the mass balance; no degradation rate calculated. 6 Oct: Do not infer cognitive group differences from separate chance-level tests; direct neuronal/matrix pathways and baseline performance remain alternatives. No established persistent benefit after withdrawal. 7 Oct: AG-D06 supplies human score association only, not TIMP2 administration. 118/119 trans variants including APOE; measured plasma levels not significantly associated with cognition. Mediation, safety and duration untested. 8 Oct: Loss of significance after removing variants can also reflect weaker protein prediction or low precision. No new sensitivity results verified; missing access is not evidence of absent analyses. AG-R05 reports direct cognitive contrasts; numerical precision and post-withdrawal benefit are not verified here. Recombinant dosing, plasma depletion and slice physiology are separate experiments; treatment units must remain distinct from nested measurements. No demonstration of selective cargo mediation. 10 Oct: Shared authorship, protocol reference, advice and disclosed interests limit attribution of independent confirmation. Overlapping animals, raw data or protein batches are not established; no blanket pooling or effect weighting. 11 Oct: The two GEO series differ in age, material and exposure. Pooling may obscure animal/litter dependence; two snRNA libraries per arm do not provide extensive independent replication. A between-series contrast isolates neither age nor therapeutic reversal of a KO effect.
Next desk check: Check the not yet fully read snRNA samples GSM9001954/55 and a WT bulk comparator. Build a small manifest of group, age, sex, batch, pool members/mouse ID, library and file; leave missing information unknown. Only then assess a bounded within-series analysis. Consider pseudobulk only for suitable cell/nucleus data; do not construct a joint genotype-by-treatment analysis.
Sources: AG-R01 · 2026-08-12; AG-R05 · 2017-04-19; AG-R08 · 2023-06-15

AG-D06 · 2025-10-23 · human association, no intervention

AG-A04 · Targeted FTL1 regulation

Preclinical · Project level 1

Outcome / target: Target: reliable neuronal function.
Population / model: Mouse.
Evidence: Mechanistic candidate.
Limits / counterevidence: Avoid blanket iron suppression; transferability unresolved.
Next desk check: Check cellular specificity, exposure and separate functional replication.
Sources: AG-S09 · 2025-08-19

AG-A05 · Redox-dependent protein control

Preclinical · Project level 1

Outcome / target: Target: function after selective DUB modulation.
Population / model: Vertebrate models.
Evidence: Mechanistic candidate.
Limits / counterevidence: Enzyme activity does not establish cognition; broad interventions ambiguous.
Next desk check: Seek selective mediation and functional counterevidence.
Sources: AG-S08 · 2026-04-21

AG-A06 · Mitochondrial quality control

Preclinical · Project level 1

Outcome / target: Target: actual restoration beyond favourable markers.
Population / model: Mouse.
Evidence: Limited; functional findings differ.
Limits / counterevidence: Do not equate prevention with reversal.
Next desk check: Prioritise age- and strain-matched comparisons.
Sources: AG-R03 · 2026-03-05

AG-A07 · Spatial control of regeneration

Preclinical · Project level 1

Outcome / target: Target: link molecular positional response, correctly patterned repeat tissue regeneration and lasting function within one system. Identical counts for different endpoints do not establish linkage.
Population / model: Axolotl development and injury regeneration; animals up to approximately eight months.
Evidence: Moderate mechanistic preclinical evidence. Cell tracing and perturbations; no established functional reversal of aging.
Limits / counterevidence: Mispatterning remains a safety criterion. Abnormal digits 4/6 occur in the revision letter without verified final-data linkage; do not equate with Shh-HCR 4/6. Cell selection limits conversion fractions; Nacu disagreement remains open. No human efficacy. AG-S03 · review record
Next desk check: Keep the axolotl linkage gap open pending suitable new source data; then compare signalling, morphology and experimental units. The planned TIMP2 group-comparison audit was completed on 6 Oct 2026 under AG-C03. AG-D05 · 2016-04-27
Sources: AG-S03 · 2025-05-21

AG-A08 · Understand senescence over time

Preclinical · Project level 1

Outcome / target: Target: repair without chronic functional harm.
Population / model: Axolotl; regeneration.
Evidence: Context-dependent mechanistic candidate.
Limits / counterevidence: No basis for indiscriminate cell removal.
Next desk check: Compare timing, cell type and state resolution.
Sources: AG-S01 · 2023-10-24

AG-A09 · AG-T01: selective cargo degradation

Hypothesis · Project level 2

Outcome / target: Hypothesis: jointly improve degradation, synapse preservation and function.
Population / model: Proposed aged neuronal system.
Evidence: Untested project synthesis.
Limits / counterevidence: Partial mediation and cargo selectivity unconfirmed. 30 Sep: the audited assays do not identify the mass balance; no degradation rate calculated. 6 Oct: Do not infer cognitive group differences from separate chance-level tests; direct neuronal/matrix pathways and baseline performance remain alternatives. No established persistent benefit after withdrawal. 7 Oct: AG-D06 supplies human score association only, not TIMP2 administration. 118/119 trans variants including APOE; measured plasma levels not significantly associated with cognition. Mediation, safety and duration untested. 8 Oct: Loss of significance after removing variants can also reflect weaker protein prediction or low precision. No new sensitivity results verified; missing access is not evidence of absent analyses. AG-R05 reports direct cognitive contrasts; numerical precision and post-withdrawal benefit are not verified here. Recombinant dosing, plasma depletion and slice physiology are separate experiments; treatment units must remain distinct from nested measurements. No demonstration of selective cargo mediation. 10 Oct: Shared authorship, protocol reference, advice and disclosed interests limit attribution of independent confirmation. Overlapping animals, raw data or protein batches are not established; no blanket pooling or effect weighting. 11 Oct: The two GEO series differ in age, material and exposure. Pooling may obscure animal/litter dependence; two snRNA libraries per arm do not provide extensive independent replication. A between-series contrast isolates neither age nor therapeutic reversal of a KO effect.
Next desk check: Check the not yet fully read snRNA samples GSM9001954/55 and a WT bulk comparator. Build a small manifest of group, age, sex, batch, pool members/mouse ID, library and file; leave missing information unknown. Only then assess a bounded within-series analysis. Consider pseudobulk only for suitable cell/nucleus data; do not construct a joint genotype-by-treatment analysis.
Sources: AG-R01 · 2026-08-12; AG-R02 · 2026-01-21; AG-R08 · 2023-06-15

AG-D06 · 2025-10-23 · human association, no intervention

AG-R05 · 19.04.2017

AG-A10 · AG-H02: FTL1–redox–DUB

Hypothesis · Project level 2

Outcome / target: Hypothesis: a causal mediation chain.
Population / model: No joint test.
Evidence: Low priority; untested.
Limits / counterevidence: Separate findings do not establish the chain.
Next desk check: First check direct prior evidence and counterevidence.
Sources: AG-S08 · 2026-04-21; AG-S09 · 2025-08-19

Updating: New approaches require a verifiable source or explicit hypothesis basis, a functional target endpoint, model limits and the next decisive countercheck. Failed or narrowed approaches remain visible with reasons. Category, evidence and project level are updated separately.

AG-A11 · Coupling antibodies to autophagy

Preclinical · Project level 1

Outcome / target: TDP-43/AAV: rotarod latency, mean difference of published values +36.46 s (7 values/group). Omitting one value per group: +28.25 to +48.88 s; not a confidence interval.
Population / model: TDP-43ΔNLS mice for motor function; separate TauP301S/sEV route for pathology findings.
Evidence: Preclinical, appraisal provisional. Published table values recalculated 1 Oct; reporting form visually inspected 2 Oct: authors report randomized mouse groups and blinding to injected samples, not independently verified.
Limits / counterevidence: Small counts, wide dispersion, experimental units/time horizon unresolved. Normal protein function and durability require appraisal; no human aging efficacy.
Next literature check: With newly accessible Jiang final full text, audit controls, experimental units, delivery, functional assay and duration. For Guo, resolve mapping of published Fig. 4e values to experimental units and timing. Revisit version metadata only when new evidence emerges.
Source: AG-D03 · 2026-08-26 · Recalculation and limits

AG-A12 · Tau antibody with an LC3-binding motif

Preclinical · Project level 1

Outcome / target: Authors report tau reduction and better motor function; final functional assay, effect size and uncertainty not ascertainable from the inspected abstract.
Population / model: P301L patient-derived neurons and JNPL3 tauopathy mice; no human intervention.
Evidence: Provisional preclinical signal; original abstract and metadata inspected. Final full text inaccessible.
Limits / counterevidence: Dose, delivery route, controls, units, duration and normal tau function unverified. Preprint probably the same study family; no double-counted replication.
Next literature check: With newly accessible Jiang final full text, audit controls, experimental units, delivery, functional assay and duration. For Guo, resolve mapping of published Fig. 4e values to experimental units and timing. Revisit version metadata only when new evidence emerges.
Source: AG-D04 · 2026-04-15 · Appraisal and limits