Antimicrobial Therapy: Drug–Bug Mechanism Design Pattern

Status: Phase 2 — Canonical Antibacterial Target-Class Module Set Complete

Six antimicrobial mechanism modules now cover the major antibacterial target classes plus the lifestyle-gating axis:

Multi-module conformers (the payoff — one disease, several independent mechanistic constraints): - Leprosy: RNA polymerase (rifampicin) + folate/DHPS (dapsone) + intracellular niche (M. leprae) - Whipple: cell wall (ceftriaxone) + ribosome (doxycycline) + folate (TMP-SMX) - Murine Typhus / Oroya Fever: ribosome (doxycycline/chloramphenicol) + intracellular niche - Leptospirosis: ribosome (doxycycline) + cell wall (penicillin/ceftriaxone)

0. Scope and Positioning (what this is and is NOT)

This work is an explanatory mechanism layer, not a drug–indication database and not a clinical decision support system (DSS):

• Not duplicating drug-indication curation. Other efforts already enumerate which drugs treat which conditions (drug labels, DrugBank, DrugCentral, ChEMBL indications, RxNorm/NDF-RT may-treat, DailyMed). dismech does not try to recapitulate that catalogue. Its contribution is to explain those associations mechanistically — linking a treatment, via target_mechanisms, to the specific pathophysiology node (drug target or gating principle) that makes it work, and to make that explanation queryable and consistency-checkable across diseases via shared modules.
• Not a DSS — claims must stay moderated. Antibiotic decision support already exists (institutional antibiograms/stewardship tools, IDSA guidelines, Sanford Guide, UpToDate, local empiric-therapy pathways, CLSI/EUCAST breakpoints). We should not position these modules as point-of-care prescribing guidance. At most this is a mechanistic substrate that such tools could cite or build on; the deliverable here is explanation and structured knowledge, not a recommendation engine.
• Open question worth a short survey: look at how existing DSSs and knowledge bases (DrugMechDB for mechanism paths; guideline/stewardship tools for drug–bug logic) represent this, to (a) borrow vocabulary and avoid reinventing, and (b) sharpen where the mechanistic-explanation niche is genuinely additive rather than overlapping. Tracked as a follow-up, not a blocker.

Recording the strategy for the treatment/action block of bacterial infectious disease entries. The same pattern generalizes to antifungal, antiparasitic, and antiviral therapy, but the worked examples here are antibacterial.

1. The Problem

Crudely, any broad-spectrum antibiotic is "good against any bacterial infection," so the naive treatment block for every bacterial entry collapses to a single generic edge: Pharmacotherapy → some antibiotic → disease. That throws away the knowledge clinicians actually use: particular mechanisms in particular antibiotics are better (or useless) for particular diseases. We want to encode that depth.

2. The Core Principle: Target the Pathogen Mechanism Node, Not the Disease

dismech already has the machinery. A Treatment links to specific pathophysiology nodes via target_mechanisms (TreatmentMechanismTarget, schema). This is the same pattern as immune_checkpoint_blockade and fibrotic_response — describe the same biology from both the disease side and the treatment side, and connect them with a target_mechanisms edge.

For antibacterials, the edge should point at the pathogen's druggable biology, which lives as pathophysiology nodes:

• Leprosy already does this for adjunctive therapy: corticosteroids → "Type 1 Delayed-Type Hypersensitivity Response"; thalidomide → "Immune-Complex Formation in Type 2 Reaction". The antibacterial agents (rifampicin, dapsone, clofazimine) currently lack target_mechanisms edges — that is the gap.
• Lyme Disease is the model for laying the substrate: it already treats pathogen biology as first-class pathophysiology nodes (Complement Evasion via BBK32, Antigenic Variation via vlsE, Host Lipid Scavenging and Membrane Biogenesis). The same entries should carry drug-target nodes (peptidoglycan synthesis, ribosomal translation, RNA polymerase, folate synthesis) that the antibiotic's target_mechanisms points at.

A single target_mechanisms edge encodes why this drug, not just that this drug — and it predicts failure modes (a cell-wall-less organism has no PBP node, so β-lactams have nothing to point at).

3. Three-Tier Encoding

Tier 1 — crude baseline (already in use). treatment_term: NCIT:C15986 (Pharmacotherapy) + therapeutic_agent (CHEBI for the specific drug, e.g. CHEBI:28077 rifampicin) + therapeutic_modality: SMALL_MOLECULE. Says "rifampicin is used for leprosy."

Tier 2 — the mechanistic edge (the depth). Add a pathophysiology node for the targeted bacterial process and link target_mechanisms to it: rifampicin → "Bacterial DNA-dependent RNA Polymerase (rpoB)"; dapsone → "Folate Synthesis (DHPS)"; a β-lactam → "Peptidoglycan Transpeptidation (PBPs)". Use target_phenotypes for adjuncts (anti-inflammatory, anti-toxin), as Leprosy already does.

Tier 3 — conserved generalizations as modules. The reasons "some mechanisms are better for some diseases" are largely conserved pathogen-property × drug-class interactions — exactly what kb/modules/ + conforms_to capture. Proposed modules: - bacterial_cell_wall_synthesis_inhibition - bacterial_protein_synthesis_inhibition - intracellular_pathogen_persistence (lifestyle gating, not a drug target)

Disease nodes conforms_to these so a generalization ("β-lactams fail vs cell-wall-less organisms") lives in one place and is queryable across diseases.

4. The Axes That Make Specific Drugs Better for Specific Diseases

Each row is a recurring reason drug choice deviates from "any broad-spectrum agent." These are the determinants worth encoding as pathophysiology nodes or agent properties.

5. Recommended Encoding Pattern (per bacterial entry)

1. Populate infectious_agent (InfectiousAgent, NCBITaxon term) — the identity anchor.
2. In pathophysiology, add drug-target nodes (cell-wall synthesis, ribosome, gyrase/topo, RNA pol, folate) and gating-property nodes (intracellular survival, toxin production, biofilm, anaerobic metabolism) as appropriate.
3. Each Treatment: Tier-1 fields plus target_mechanisms → the specific node(s) it inhibits, and target_phenotypes for adjuncts.
4. Where the mechanism recurs, conforms_to a shared antibiotic-mechanism module.

6. Open Decision for the Register

Surface to docs/explanation/design-decisions.md before building: whether drug–bug mechanism knowledge lives per-disease (treatments hang off the Disease, consistent with current architecture — recommended) or whether antibiotic classes deserve their own module/grouping objects that diseases reference. The module route (Tier 3) gives both: disease-centric storage with a conserved-mechanism layer, no new top-level class needed.

7. Next Steps

• [x] Draft a bacterial_cell_wall_synthesis_inhibition module as proof-of-concept. Built at kb/modules/bacterial_cell_wall_synthesis_inhibition.yaml: five nodes (precursor/lipid II synthesis → PBP cross-linking → bactericidal autolysis, plus acquired-resistance and intrinsic cell-wall-deficient gating branches). Schema + term validation pass; all evidence snippets are verified exact substrings of real abstracts (Typas 22203377, Sauvage 18266856, Blair 25435309, Pereyre/Tardy 34680797, Kim 18302341). Key conformance/treatment target: #Peptidoglycan Cross-Linking by Penicillin-Binding Proteins.
• [x] Wire conforming entries end-to-end. Ten bacterial-disease entries now carry a drug-target pathophysiology node with conforms_to the module plus a treatment target_mechanisms edge to that node:
  • β-lactam → PBP cross-linking node: Lyme Disease (amoxicillin, ceftriaxone), Scarlet Fever (penicillin), Bejel + Pinta (benzylpenicillin), Whipple Disease (ceftriaxone), Paratyphoid Fever (ceftriaxone), Ludwigs Angina (penicillin), Bacterial meningitis (ceftriaxone).
  • glycopeptide → lipid II / precursor node: Clostridioides_difficile (oral vancomycin), Bacterial meningitis (vancomycin, second node).
  • acquired-resistance node: Furunculosis models both the β-lactam target (oxacillin/MSSA) and the Acquired Resistance and Drug Inactivation node (PBP2a/MRSA → vancomycin substitution), the clearest drug-choice gating case. All ten pass schema + term validation.
• [ ] Deliberately excluded (primary therapy is not a cell-wall agent, so they do not conform to this module): Leptospirosis, Yaws (doxycycline/azithromycin first-line), Tetanus (metronidazole + antitoxin; β-lactam adjunct only), Oroya Fever (intracellular Bartonella), Folliculitis (often topical/mixed). These are candidates for the future intracellular_pathogen_persistence and protein-synthesis-inhibitor modules instead.
• [x] Draft bacterial_protein_synthesis_inhibition module. Built at kb/modules/bacterial_protein_synthesis_inhibition.yaml: three nodes (ribosomal translation target → toxin/exoprotein-synthesis suppression → ribosomal target resistance). Evidence: Wilson 24336183 (ribosome target + resistance), Sawai 17101685 (clindamycin exoprotein suppression). Schema + term validation pass.
• [x] Draft intracellular_pathogen_persistence module (lifestyle-gating). Built at kb/modules/intracellular_pathogen_persistence.yaml: two nodes (intracellular niche / beta-lactam exclusion → cell-penetrant-drug requirement). Evidence: Maurin & Raoult 18611821, Pea 28639230. Schema + term validation pass.
• [x] Multi-module conformers wired: Murine Typhus (doxycycline) and Oroya Fever (chloramphenicol) each conform to BOTH the ribosome target (#Bacterial mRNA Translation by the Ribosome) and the intracellular gating node (#Intracellular Niche and Beta-Lactam Exclusion), demonstrating composition of a target-based module with a pharmacokinetic-gating module.
• [x] Built the remaining target-class modules: bacterial_dna_topoisomerase_inhibition (fluoroquinolone), bacterial_rna_polymerase_inhibition (rifamycin), and bacterial_folate_synthesis_inhibition (sulfonamide/trimethoprim/dapsone). Evidence: Hooper & Jacoby 27449972, Mosaei & Zenkin 32342856, Capasso & Supuran 23627736, Bourne 27025730. All pass schema + term validation.
• [x] Wired remaining ribosome conformers (Lyme, Leptospirosis, Yaws, Whipple, Scarlet), plus fluoroquinolone (Travelers' Diarrhea), rifamycin (Leprosy, Buruli), folate (Leprosy, Whipple), and a fixed cell-wall link on Whipple (its earlier node had no treatment edge — now corrected).
• [ ] The #Suppression of Toxin and Exoprotein Synthesis node has no conformer yet — needs a toxin-mediated entry (necrotizing fasciitis / strep or staph toxic shock), which does not currently exist in kb/disorders/.
• [ ] Short survey of existing drug-mechanism KBs / DSSs (DrugMechDB, stewardship tooling) — see §0; borrow vocabulary, sharpen the explanatory niche.
• [ ] Record the per-disease-vs-module decision in the design register.