Drugs With Unknown or Contested Mechanisms of Action

Why this matters for dismech

A treatments[].target_mechanisms edge says this drug acts on this node of the causal graph. For most drugs that is a well-characterized claim. For a set of widely used drugs (valproate, lamotrigine, topiramate, metformin, lithium, ketamine, cannabidiol, paracetamol, modafinil, riluzole) the mechanism that produces the therapeutic effect is still unknown or disputed, even though the clinical effect is not. On those, a target_mechanisms edge can state a mechanism more confidently than its source does, and the schema has no way to record that the mechanism is open.

Source list

Wikipedia's Category:Drugs with unknown mechanisms of action (154 pages and a General anesthetics subcategory, read 2026-10-12) is the seed list. It is a lead, not a ruling:

• Membership is editor-assigned and goes stale. Lenalidomide is listed, although cereblon-dependent degradation of IKZF1/IKZF3 is now the accepted mechanism. Clavulanic acid is listed, although it is a beta-lactamase inhibitor.
• "Unknown" refers to the drug's best-known use. Ethanol is listed for its psychoactive effects. As a toxic-alcohol antidote, competitive saturation of alcohol dehydrogenase is well understood.
• Most of the 154 pages are research compounds, racetams, and modafinil analogues that the KB does not use.

Overlap with the KB

String matching on treatment name and treatment_term (agents included) across kb/disorders, kb/modules and kb/comorbidities found 25 of the listed drugs in 248 treatment records across 193 entries. 113 of those records carry target_mechanisms (in 93 entries). The full table is DRUGS_UNKNOWN_MECHANISM/kb_treatment_overlap.tsv; the discussions already attached to each record are in kb_treatment_discussions.tsv. Matching is lexical, so a multi-drug treatment ("Valproate with lamotrigine") is counted once per drug.

Triage

The 113 linked records fall into three groups. Only the third needs review.

1. Symptomatic targets make no mechanism claim. An antiseizure medication targeting Seizures or Cortical Network Hyperexcitability says the drug suppresses the phenotype, which is the documented clinical effect. The schema explicitly allows this. Most valproate, lamotrigine, topiramate and cannabidiol records are here. No action needed.
2. The mechanism in this use is known, despite the Wikipedia listing. Ethanol as an ADH-competing antidote (Ethylene_Glycol_Poisoning, Methanol_Poisoning, Diethylene_Glycol_Poisoning, toxic_alcohol_adh_bioactivation); selegiline and rasagiline as MAO-B inhibitors in Parkinsons_Disease; memantine as an NMDA channel blocker on GRIN gain-of-function nodes; clavulanic acid in Pasteurellosis. No action needed beyond checking the evidence.
3. A specific molecular mechanism is asserted for a drug whose therapeutic mechanism is contested. These are the worklist: - Metformin, six aging modules. Each module asserts a different mechanism: AMPK/mTORC1 (deregulated_nutrient_sensing), autophagic flux (disabled_macroautophagy), microbiome (gut_dysbiosis), anti-inflammatory (inflammaging), mitochondrial (mitochondrial_dysfunction), stem-cell (stem_cell_exhaustion). These may all be true, but read together they show that no single mechanism has been established. Also check Type_2_Diabetes_Mellitus (hepatic glucose output, where the molecular route is contested), Lafora_Disease, Maple_Syrup_Urine_Disease, Li-Fraumeni_Syndrome (already covered by a gap, see below), Classic_Familial_Adenomatous_Polyposis, Oral_Cavity_Squamous_Cell_Carcinoma. - Lithium in Bipolar_Disorder. Four parallel targets (GSK-3β, phosphoinositide, circadian, neuroplasticity). These are the competing hypotheses, not an established mechanism. They are a candidate for mechanistic_hypotheses groups. - Valproate as an HDAC inhibitor (BRPF1-Related_Intellectual_Disability, and Li-Ghorbani-Weisz-Hubshman_Syndrome, which is already covered by a gap) and on mitochondrial nodes (Autosomal_Recessive_Spinocerebellar_Ataxia_20). HDAC inhibition is a real property of valproate, so check that the evidence shows it acting through that route in this disease. - Ketamine in Major_Depressive_Disorder and Post-Traumatic_Stress_Disorder (the antidepressant mechanism beyond NMDA block is disputed), modafinil in Idiopathic_Hypersomnia, topiramate in pseudotumor_cerebri (CSF hypersecretion), lenalidomide in Immunodeficiency_64 (already covered by a gap), paracetamol in Patent_Ductus_Arteriosus, riluzole in Friedreich_Ataxia and Spinocerebellar_Ataxia_Type_1, edaravone in Amyotrophic_Lateral_Sclerosis.

For each record in group 3, the review is:

• Read the target_mechanisms[].evidence. If the quote shows a clinical response and the mechanism is inferred from it, mark the item directness: INDIRECT. That is what INDIRECT means: a therapeutic response cited as validation of the mechanism it targets.
• If the entry carries competing mechanisms for one drug, consider expressing them as mechanistic_hypotheses with hypothesis_groups on the edges rather than as parallel settled edges.
• If no evidence supports the specific node, retarget the edge to the phenotype the drug treats, or remove it. Do not add evidence in order to keep the edge.
• Record the open mechanism as a KNOWLEDGE_GAP discussion (next section).

Recording an unknown mechanism: a knowledge gap on the treatment

No new slot is needed. A discussions entry with kind: KNOWLEDGE_GAP whose attaches_to names both the treatment and the node its edge targets says exactly this: the drug works, and how it works here is not established. The prompt states the question, the rationale says which candidate mechanisms the evidence cannot tell apart, and proposed_experiments can say how they could be.

Three entries already do this and are the pattern to copy:

HNRNPU-Related_Developmental_and_Epileptic_Encephalopathy asks a similar question about valproate (gap_valproate_efficacy_mechanism) but is not the pattern: it is an OPEN_QUESTION attached only to pathophysiology#Recurrent Seizures, not to a treatment.

Coverage

Of the 113 linked records, 47 have some discussion attached to the treatment or to a node it targets. Almost all of those discussions are about the node, for example whether a biomarker tracks aging. They are not about the drug. Apart from the three pattern entries above (Li-Fraumeni_Syndrome, Li-Ghorbani-Weisz-Hubshman_Syndrome, Immunodeficiency_64), none of the group 3 records had a gap about the drug's own mechanism when this page was written. That included the metformin module edges, lithium in Bipolar_Disorder and both ketamine psychiatric entries. Adding one per record, attached to the treatment and its target, is the main deliverable of this project.

Rules for writing them:

• One discussion per entry and drug, attached to every node the drug targets there. Do not create one per edge.
• The rationale names the candidate mechanisms and says why the cited evidence does not discriminate between them. It needs its own evidence, and ideally a source that says the mechanism is unclear in so many words, as the MILI protocol does for Li-Fraumeni. A Wikipedia category listing does not count.
• A gap does not replace fixing an edge that overstates its evidence. Do both: set directness: INDIRECT where the mechanism is inferred from a clinical response, and record the gap.
• Where the literature argues for two or more specific models (lithium), a CONTROVERSY discussion or mechanistic_hypotheses groups may fit better than KNOWLEDGE_GAP. Use KNOWLEDGE_GAP when the honest answer is that nobody has tested it.

Progress

Everything listed as done below is in PR #14110 and is not on main until that PR merges.

Done (one KNOWLEDGE_GAP each, attached to the treatment and its targets): deregulated_nutrient_sensing, disabled_macroautophagy, gut_dysbiosis, inflammaging, mitochondrial_dysfunction (metformin) and Bipolar_Disorder (lithium, gap_bipolar_lithium_therapeutic_mechanism).

Not done, on purpose: stem_cell_exhaustion. The study behind that metformin edge (PMID:33524231) tested the mechanism: siRNA knockdown of AMPKα abolished metformin's effect on NF-κB activation and the senescence-associated secretory phenotype in cultured CKD mesenchymal stem cells. The mechanism in that system is therefore supported, and an "unknown mechanism" gap would be false. What remains open is whether ex vivo preconditioning says anything about systemic metformin and resident stem cells in aging tissue. That is a translational question, not this project's.

Done with an OpenScientist report behind each (just research-drug-mechanism, reports in research/drug_mechanism/). Every quote used was re-checked against references_cache/:

The reports are leads, not evidence. Two of their claims did not survive checking against the cited source and were left out (see the history records). OpenScientist jobs can exceed the client's default one-hour timeout; pass --param timeout=7000.

Still pending for all ten records: the directness: INDIRECT check on the existing edge evidence. The gaps were added without changing any edge, and several edges carry no evidence of their own, including every metformin module edge and the lithium circadian edge.

The knowledge gaps and research reports themselves are in PR #14110; this page, the recipe and the template are kept separate because the curation scope check does not allow tooling in a KB PR.

Remaining: the other group 3 records above.

A schema change, such as a treatment-level "mechanism unknown" flag, would duplicate this and is not proposed.

Reproducing the overlap

The table was built by matching drug names (with common synonyms: acetaminophen, mesalamine, valproic, dipyrone) against each treatment's name and treatment_term, excluding evidence text. It is not a gate and should not become one. Membership in the Wikipedia category is not a property of a KB treatment.