Dementia Anchor Triage: A Public‑Data, Evidence‑Led Demo

Auspexi • Updated:

TL;DR: We outline a public‑data methodology to turn well‑known anchors (genetics, expression, pathways, chemistry, trials) into a ranked, explainable shortlist of targets and repurposable compounds. The focus is on reproducible, signed outputs—not PHI, not proprietary cohorts. This demonstrates practical capability and an investable path for funded pilots.

What we are demonstrating (and what we are not)

Demonstrating: A small, verifiable workflow that ingests public anchors, scores candidates, and produces human‑readable rationales with citations and signed run metrics.
Not claiming: We are not running clinical research, and we are not making treatment claims. This is a capability demo to accelerate literature‑driven triage and reviews.

Anchors (public‑first)

Genetics: GWAS hits and rare variants (e.g., IGAP) — locus support and colocalization.
Expression: bulk/single‑cell brain datasets (e.g., AMP‑AD/ROSMAP, Allen/HCA) with region/cell specificity.
Pathways: Reactome/WikiPathways (microglial/innate immune, autophagy‑lysosome, mitochondrial, synaptic).
Chemistry: Open Targets, ChEMBL, PubChem for ligandability and selectivity; PDB for structures.
Trials & safety: ClinicalTrials.gov, SIDER/FAERS (signals, contraindications).

Methodology (plain English)

Ingest & normalize public anchors into a provenance graph with licenses and citations.
Score targets by a transparent formula combining genetic support, brain‑cell expression direction, and pathway centrality.
Score compounds for tractability (activity proxies), brain exposure proxies, and safety signals.
Shortlist 3–5 targets with 1–2 compounds each; generate a one‑pager per candidate with citations and “next step.”
Sign outputs: notebooks print a sha256 of inputs/outputs for reproducibility; versions are recorded.

What a viewer sees in the demo

A Target Triage notebook producing a ranked list (e.g., TREM2, SORL1, CD33) with reasons.
A Compound Triage notebook with a small shortlist and risk notes (activity, BBB proxy, safety flags).
Two one‑pagers (target + compound) that read like a mini briefing with links.
A benchmark JSON with run timestamp and hash (“signed metrics”).

Why this matters commercially

Faster reviews: move from discussion to a small, shared artifact with citations.
Lower risk: signed runs and transparent scoring reduce ambiguity.
Pilot‑ready: with funding and partners, swap public anchors for controlled datasets and add wet‑lab loops.

How to run our scaffold

Target notebook: notebooks/dementia_target_triage.ipynb
Compound notebook: notebooks/dementia_compound_triage.ipynb
Templates: docs/templates/target_one_pager.md, docs/templates/compound_one_pager.md

Slide deck (investor/customer)

Preview the 10‑slide outline: /slides/dementia-demo/

Ethics & compliance

No PHI: public anchors only; controlled‑access data appear only under appropriate approvals.
Reproducibility: signed runs, versioned inputs, deterministic re‑runs.
Clear scope: this is a triage and communication tool, not a claim of efficacy.

Request a Pilot • Contact Sales