An ontology-driven knowledge graph approach to fibromyalgia research — every claim carries an explicit confidence tier, every mechanism link traces back to evidence, and bridges to non-FM literature are first-class research objects rather than digressions.
The standard literature review distills consensus. That is useful — but consensus alone doesn't produce a cure. To find a cure, you have to identify the load-bearing claims (the ones that, if right, change everything) and the load-bearing experiments (the ones that, if successful, close the question). And you have to do both while distinguishing what's established from what's plausible-but-unproven.
This project's structural commitment is to make that distinction visible in every claim. Each mechanism link in the knowledge graph carries an explicit confidence tier. Each open question carries an explicit status. Promoting a claim from one tier to a higher one is a deliberate, logged act backed by triggering evidence — never a silent re-write.
The outputs the project aims for are not patient-advice content. They are foundational scientific paths toward a cure: identified mechanism nodes, tractable trial designs, and biomarker stratifiers that could route patients to subtype-specific therapy.
A claim's tier determines how much weight it carries in synthesis and which experiments would move it up or down the ladder.
Multiple replicated peer-reviewed studies. The floor of consensus — these are findings the field has agreed on.
Example: Central sensitization with reduced descending inhibition. Reduced IENFD in 40–60% of FM patients. CSF substance P elevation.
Peer-reviewed, but not yet replicated at scale. Most active mechanism research operates at this tier. Promotion to established requires independent replication in well-powered cohorts.
Example: FM-IgG passive transfer reproducing pain in mice (Goebel 2021, replicated). Anti-SGC IgG in 37% of FM patients (Seefried 2025).
Synthesized across multiple papers — not directly stated in any single source. Held with deliberate caution; explicitly labeled in the synthesis files so downstream readers can audit the inference.
Example: Triangulation that FM has at least three mechanistically distinct subtypes (SFN+, autoimmune+, inflammatory-central+).
Hypotheses drawn from non-FM literature. Speculative but mechanistically motivated. Promotion to emerging requires direct demonstration in FM tissue or FM patients.
Example: Lipedema-FM comorbidity via MC-estrogen-receptor mechanism (35–40% comorbidity, Amato 2026 framework).
Each tier promotion gets a row in the promotion log with the triggering evidence: paper(s), date, what changed, and net effect on the synthesis. The 2026-05-21 ingestion of Sanchez 2025, for instance, promoted bridge B1 (long COVID ↔ FM via autoantibody-mediated SFN) from bridging to emerging because both endpoints of the bridge now have direct passive-transfer evidence in mice.
This is not a polite footnote. It is the system's mechanism for preventing silent drift.
The project organizes its core mechanistic hypotheses into three numbered chains. They are not exhaustive, but they capture the high-leverage research surfaces. Each chain has its own end-to-end mechanism, its own biomarker signature, and its own candidate intervention point.
Gut dysbiosis → microbially-produced bile acids → B-cell / plasma-cell dysregulation → FM-IgG → mast cell activation (via MRGPRX2) → sensitization.
Anchor evidence: Goebel 2021 (passive transfer); Krock 2023 (titer↔severity); Seefried 2025 (37% positive); Jakobsson 2026 (bile acids); Sanchez 2025 (MRGPRX2 receptor identification).
Multiple ligand inputs (FM-IgG, substance P, estrogen at gfWAT) converge on mast cell MRGPRX2 → degranulation → IL-6/histamine/tryptase → peripheral sensitization. Heritable amplifier: HαT/TPSAB1 duplication.
Anchor evidence: Lyons 2019 (HαT); Christoforou 2026 (cromolyn pilot); Aitella 2026 (MC-SP loop); Simeone 2026 (mucosal MRGPRX2); Amato 2026 (lipedema framework).
Disrupted Bayesian precision-weighting in interoceptive hierarchy → reduced body-state-map differentiation → alexithymia + brain fog + cognitive symptoms. Network substrate: thalamocortical decoupling.
Anchor evidence: Strube 2026 (predictive-coding framework); Herman 2026 (emBODY task); Hou 2026 (thalamocortical decoupling); Kerrebijn 2025 (FM GWAS brain-tissue enrichment).
The 2025-2026 development that this project tracks most closely: H1 and H2 are no longer parallel. Sanchez 2025 unifies them at the MRGPRX2 receptor, with FM-IgG and substance P converging on the same effector cell. H3 remains structurally distinct — Mignolet 2026 long-COVID IgG passive transfer produces sensory pain (which transfers via the IgG mechanism) but not cognitive symptoms (which require a separate axis).
This kind of structural finding — what is a single mechanism vs what are parallel mechanisms — is precisely what the ontology-driven approach is built to detect.
A traditional review treats outside-the-field findings as a digression. This project treats them as bridge candidates — explicit, named connections between FM-side entities and non-FM-side entities that may carry mechanism between them.
Every bridge is defined by its shared entities (what does FM share with the other domain?), its motivating mechanism (why might the link matter?), and its closing condition (what evidence would either promote it to an established cross-condition mechanism or refute it as an over-reach?).
The project tracks ten active bridges as of 2026-05-21. Each has its own evidentiary history and its own next experiment.
Bridges with shared entities across three or more domains get higher priority than those across only two. New bridges should be considered any time a paper from outside the FM literature is read — that openness is the project's primary mechanism for finding non-obvious mechanism connections.
The project includes a small Python agent that queries Europe PMC using ontology-derived bridge queries. The agent never auto-ingests — it stages candidates, and a human (or AI assistant operating under explicit instructions) runs the full intake protocol.
The intake protocol has six steps: (1) read for entities and relationships, (2) reconcile with the ontology, (3) create a paper file with claims-as-triples, (4) update the master index, (5) update affected synthesis files, (6) look for triangulation across existing papers. Skipping any of these would compromise the ability to trace claims back to evidence — which is the system's central capability.
As of 2026-05-21 the agent has a maturity-aware filter: bridges that have reached flagship-anchor status (B2 microbiome, generic long COVID, metformin cross-condition) are placed in cooldown for 2-3 months, suppressing duplicate review-tier yield. Reopens when a fresh mechanism input or primary cohort study justifies it. This is documented in synthesis/agent-tuning-2026-05-21.md with reopening rules.
Three cure-path arms have crossed from speculative into clinically tractable in the last twelve months. Each addresses a specific mechanism link in the H1×H2 unified chain. Each is anchored on either an FDA-approved compound or a Phase 2/3 clinical-stage compound.
These are not the only paths the project tracks — there are 84 active open questions and 36 active bridges. But these three are the ones where the mechanism case is strongest and the clinical-stage tooling is closest to ready.