Version history

{
  "session_id": "sess_SDA-2026-04-25-gapdebate-98a600b3ed_c747c608",
  "round_number": 3,
  "agent_persona": "persona-domain_expert",
  "agent_backend": "codex_cli",
  "action": "assess",
  "content": "Most investable survivors are `6 > 4 > 1 > 2`. I would not spend serious translational budget yet on `7`, and I would treat `3` and `5` as modifier mechanisms rather than lead programs.\n\n| Rank | Hypothesis | Druggability | Biomarkers | Best model systems | Safety / translational risk | Realistic path |\n|---|---|---|---|---|---|---|\n| 1 | `6` Tau missorting transitions into autonomous tau seeding | High, relative to others. Clear intervention classes: anti-tau antibodies, seed-blocking biologics, ASOs, uptake blockers. | CSF/plasma p-tau217, p-tau181, MTBR-tau, tau seeding assays, tau PET, synaptic markers like NfL/neurogranin. | Human iPSC excitatory neuron networks, microfluidic compartments, seeded organoids, APP+tTAU mouse combinations with amyloid withdrawal. | Main issue is timing: likely only works early. Anti-tau antibodies may show modest efficacy if seeds are mostly intracellular. | Strong preclinical package possible in 18-30 months; IND-grade program 3-5 years; roughly `$15M-$40M` for serious preclinical-to-IND effort depending on modality. |\n| 2 | `4` Microglia/complement sustain post-Aβ degeneration | Moderate. Targets exist: C1q, C3, CR3, TREM2-state modulators. Biology is druggable, but CNS immunology is tricky. | CSF/plasma YKL-40, sTREM2, C1q/C3 fragments, SV2A PET for synapses, tau PET, NfL. | Human tri-cultures, xenografted human microglia mice, amyloid-plus-tau models with delayed amyloid clearance, spatial transcriptomics. | Biggest risk is on-target immunologic liability and blocking beneficial pruning/repair. Likely better for slowing synapse loss than reversing tau polarity. | Mechanistically testable in 12-24 months; translational de-risking 2-4 years; `$10M-$25M` for platform-quality preclinical program. |\n| 3 | `1` Fyn-anchored dendritic tau/NMDAR complex persists after Aβ | Moderate. Fyn is druggable, but prior CNS kinase efforts have struggled. Tau-lowering may be better than direct Fyn inhibition. | Phospho-NMDAR/SRC signatures in CSF are weak clinically; better to pair CSF tau markers with EEG/network hyperexcitability, SV2A PET, synaptic fluid biomarkers. | Primary neurons are useful, but human iPSC neuron-astrocyte co-cultures and MEA readouts are more relevant; inducible Aβ pulse-washout paradigms are essential. | Narrow therapeutic window, CNS kinase tolerability, and risk that rescue only reflects generic anti-excitotoxic effects. | Good mechanistic program in 12-18 months; hard to make into differentiated clinical asset unless paired with tau-lowering; `$8M-$20M` preclinically. |\n| 4 | `2` GSK3β/CDK5 feedback maintains missorting | Low-moderate. Targets are druggable in principle, but specificity and therapeutic index are poor. Best use is mechanistic validation, not near-term drug program. | p-tau epitope panels, kinase activity proxies, phosphoproteomics, NfL. | Compartmentalized neurons, live-cell phospho-tau imaging, inducible kinase perturbation after washout. | Pleiotropy is the problem. GSK3β/CDK5 inhibition affects many essential functions, so safety margin is likely weak for chronic AD use. | Fast falsification in 9-15 months; not attractive as a standalone development thesis unless it points to a more specific downstream node; `$3M-$8M` for decisive mechanistic work. |\n\nMy call on the others:\n\n- `3` Proteostasis failure is biologically credible but too nonspecific for first-pass translation. Good as a downstream readout, weak as a lead therapeutic thesis.\n- `5` Astrocytic EAAT2 failure is plausible and may matter in subgroups with hyperexcitability, but it is unlikely to be the core persistence mechanism.\n- `7` BIN1 should be treated as a genetic modifier study, not a drug program.\n\nBest biomarker package across the whole question:\n- Entry/target engagement: amyloid PET or soluble Aβ measures proving true clearance.\n- Persistence biology: CSF/plasma p-tau217, MTBR-tau, tau seeding assay.\n- Damage state: NfL, neurogranin, SV2A PET.\n- Mechanism-specific add-ons:\n  - `4`: sTREM2, C1q/C3, microglial imaging/transcriptomics.\n  - `1`: EEG/MEA hyperexcitability, calcium imaging.\n  - `2`: phosphoproteomics.\n  - `6`: biosensor seeding assays from conditioned media and cell fractions.\n\nBest experimental strategy:\n1. Build one rigorous `pulse Aβ -> verified complete removal -> delayed intervention` assay.\n2. Run it first in human iPSC neuron monoculture and tri-culture.\n3. Prioritize endpoints that separate `tau polarity`, `tau seeding`, and `ongoing degeneration`.\n4. Advance only mechanisms that remain abnormal after orthogonally proving near-zero residual Aβ.\n\nIf you want, I can turn this into a SciDEX-ready table with `tractability_score`, `trial_readiness_score`, `killer_experiment`, `timeline_years`, and `budget_range_usd`.",
  "tokens_used": "1188",
  "persona_id": "persona-domain_expert"
}