Antiviral Therapy: Drug–Virus Mechanism Design Pattern

Status: Phase 1 — Antiviral Mechanism Module Set Built

This project extends the ANTIMICROBIAL drug–bug mechanism design pattern to antiviral therapy. The same machinery applies — a Treatment links via target_mechanisms to the specific pathophysiology node (a viral enzyme/step or a gating principle) that makes the drug work, and recurrent virus-property × drug-class interactions are captured once as kb/modules/ that disease entries conforms_to.

Six antiviral mechanism modules are now built and validated (schema + term + independent snippet-substring verification of every evidence quote): five direct-acting viral-target modules plus the latency/reservoir gating module. The remaining work is wiring conforming disease entries (conforms_to + treatment target_mechanisms edges), tracked in §7.

Multi-module conformers (the payoff — one disease constrained by several independent antiviral mechanisms, mirroring Leprosy/Whipple on the antibacterial side): - Acquired Immunodeficiency Syndrome (HIV/ART): reverse transcriptase (viral_polymerase_inhibition) + integrase (viral_integrase_inhibition) + protease (viral_protease_inhibition) + entry (viral_entry_fusion_inhibition) + latent reservoir (viral_latency_reservoir_persistence). The canonical combination-therapy story. - Hepatitis C (direct-acting antiviral regimens): NS5B polymerase (viral_polymerase_inhibition) + NS3/4A protease (viral_protease_inhibition) + NS5A replication-complex (viral_assembly_release_inhibition). The three-target antiviral cure. - Hepatitis B: reverse transcriptase suppression (viral_polymerase_inhibition) gated by the cccDNA reservoir (viral_latency_reservoir_persistence) — explains indefinite suppression without cure.

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

This is an explanatory mechanism layer, not a drug–indication database and not a clinical decision support system (DSS). Identical positioning to ANTIMICROBIAL §0:

• Not duplicating drug-indication curation. DrugBank, DrugCentral, ChEMBL, RxNorm/NDF-RT, and antiviral drug labels already enumerate which antivirals treat which infections. dismech's contribution is to explain those associations mechanistically — linking a treatment, via target_mechanisms, to the viral step it blocks — and to make that explanation queryable and consistency-checkable across diseases via shared modules.
• Not a DSS. HIV/HCV/HBV treatment guidelines (DHHS, AASLD/IDSA, EASL), resistance-interpretation tools (Stanford HIVdb), and stewardship pathways already give point-of-care guidance. These modules are at most a mechanistic substrate such tools could cite, not a prescribing engine.
• Open question worth a short survey: how DrugMechDB and antiviral resistance databases represent viral drug-target mechanism paths — borrow vocabulary, avoid reinventing, and sharpen where mechanistic explanation is additive. Tracked as a follow-up, not a blocker.

1. The Problem

Naively, the treatment block for every viral entry collapses to Pharmacotherapy → some antiviral → disease. That discards the knowledge clinicians actually use: viruses are far more target-divergent than bacteria — there is no broad-spectrum antiviral analogous to a broad-spectrum antibiotic, because each virus family carries its own polymerase, protease, and entry machinery. Which mechanism a drug hits, and whether that target even exists in a given virus, is the whole game. We want to encode that depth.

2. The Core Principle: Target the Viral Step Node, Not the Disease

A Treatment links to specific pathophysiology nodes via target_mechanisms (TreatmentMechanismTarget). For antivirals, the edge should point at the virus's druggable step, which lives as a pathophysiology node:

• Acute Hepatitis C Virus Infection is the model already laid down: its direct-acting antiviral treatment carries a target_mechanisms edge (INHIBITS) onto an "Early HCV infection of hepatocytes" node and explains that the NS5B polymerase inhibitor sofosbuvir plus NS5A inhibitor velpatasvir "suppress HCV replication during the early-infection phase." That is the target — a finer-grained "NS5B RNA-Dependent RNA Polymerase" node would let the polymerase module attach.
• COVID-19 already carries the host/viral-evasion biology as first-class pathophysiology and conforms_to the parp_parg_macrodomain_viral_evasion module, and lists nirmatrelvir (Mpro inhibitor) and remdesivir (RdRp inhibitor) as treatments — but without target_mechanisms edges onto explicit "SARS-CoV-2 Main Protease (Mpro/3CLpro)" and "SARS-CoV-2 RNA-Dependent RNA Polymerase" drug-target nodes. That is the gap to close and the cleanest first proof-of-concept.

A single target_mechanisms edge encodes why this drug, and it predicts failure modes: a DNA-virus polymerase inhibitor has nothing to point at in an RNA virus, and an integrase inhibitor is meaningless outside the retroviruses.

3. Three-Tier Encoding

Tier 1 — crude baseline. treatment_term: NCIT:C16119 (Antiviral Therapy therapy) or NCIT:C15986 (Pharmacotherapy) + therapeutic_agent (CHEBI for the drug, e.g. CHEBI:85083 sofosbuvir, CHEBI:145994 remdesivir, CHEBI:170007 nirmatrelvir) + therapeutic_modality: SMALL_MOLECULE. Says "remdesivir is used for COVID-19."

Tier 2 — the mechanistic edge (the depth). Add a pathophysiology node for the targeted viral step and link target_mechanisms to it: remdesivir → "SARS-CoV-2 RdRp"; nirmatrelvir → "SARS-CoV-2 Mpro"; sofosbuvir → "HCV NS5B polymerase"; dolutegravir → "HIV integrase strand transfer"; oseltamivir → "Influenza neuraminidase-dependent virion release". Use target_phenotypes for host-directed adjuncts (e.g. dexamethasone → COVID hyperinflammation), exactly as the antibacterial entries do for anti-inflammatory adjuncts.

Tier 3 — conserved generalizations as modules. The reasons "some mechanisms are better/useless for some viruses" are conserved virus-property × drug-class interactions — kb/modules/ + conforms_to. See the proposed module table above. The host-side parp_parg_macrodomain_viral_evasion module already exists and is complementary: it captures the host antiviral / viral-evasion axis (IFN-induced PARP ADP-ribosylation vs viral macrodomain countermeasures), whereas the proposed modules capture the direct antiviral drug-target axis.

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

5. Recommended Encoding Pattern (per viral entry)

1. Populate infectious_agent (InfectiousAgent, NCBITaxon term) — the virus identity anchor.
2. In pathophysiology, add viral-step drug-target nodes (polymerase, protease, entry/fusion, integrase, assembly/release) and gating-property nodes (latency/reservoir, quasispecies/resistance, compartment sanctuary) as appropriate.
3. Each Treatment: Tier-1 fields plus target_mechanisms → the specific node(s) it inhibits, and target_phenotypes for host-directed adjuncts.
4. Where the mechanism recurs, conforms_to a shared antiviral-mechanism module.
5. Keep direct-acting antivirals distinct from host-directed therapy — they target different node types (viral step vs host phenotype).

6. Open Decision for the Register

Surface to docs/explanation/design-decisions.md: the same per-disease-vs-module decision recorded for antibacterials applies here, plus an antiviral-specific question — whether host-directed antiviral mechanisms (IFN signaling, restriction factors, the existing parp_parg_macrodomain_viral_evasion module) should be modeled as the same kind of object as direct-acting viral-target modules, or kept as a distinct host-axis layer. Recommendation: keep both as kb/modules/, but tag direct-acting (virus-target) vs host-directed so queries can separate them.

7. Next Steps

• [x] Draft viral_polymerase_inhibition as the proof-of-concept module (nucleos(t)ide chain termination + non-nucleoside allosteric block across RdRp / reverse transcriptase / viral DNA polymerase), mirroring the bacterial_cell_wall_synthesis_inhibition build. Built at kb/modules/viral_polymerase_inhibition.yaml: four nodes (polymerase target → chain-termination/replication-arrest → non-nucleoside allosteric inhibition → quasispecies-driven target resistance). Evidence: Geraghty 33924302 (broad-spectrum nucleoside analogues, remdesivir delayed chain termination, sofosbuvir prodrug, high-mutation-rate escape), Amblard 35792384 (HIV NRTIs). Key target: #Viral Polymerase as the Replication Engine and Drug Target.
• [x] Draft viral_protease_inhibition. Built as a shared polyprotein-synthesis and virus-encoded protease-processing trunk with separate replicase-protein/RNA-replication (SARS-CoV-2, HCV) and structural virion-maturation (HIV) outputs, plus a viral target-resistance branch. The immature non-infectious-particle consequence is confined to the structural branch, and pharmacokinetic boosting is separate from viral resistance. Evidence: 32858867 (HIV maturation), 35380892 (SARS-CoV-2 replicase-polyprotein cleavage), 22558217 (HCV NS3/4A polyprotein processing), and 35993498 (HCV NS3/4A resistance). Key target: #Virus-Encoded Protease-Dependent Polyprotein Processing.
• [x] Draft viral_entry_fusion_inhibition (attachment / co-receptor / fusion / NTCP), viral_integrase_inhibition (retroviral strand transfer), and viral_assembly_release_inhibition (influenza neuraminidase + HCV NS5A; baloxavir deliberately excluded as a PA-endonuclease target). All built, each with a target → consequence → resistance/gating node set and verified evidence.
• [x] Draft viral_latency_reservoir_persistence (gating, not a drug target — the antiviral analog of intracellular_pathogen_persistence). Built: three nodes (latent/archival genome → suppression-without-eradication → reactivation on interruption). Evidence covers the HIV proviral reservoir, HBV cccDNA, and HSV neuronal latency. Key gating target: #Antiviral Suppression Without Eradication of the Reservoir.
• [ ] Wire remaining conforming disease entries. The viral_protease_inhibition replicase branch is now instantiated in COVID-19 (pp1a/pp1ab → Mpro processing → replication-transcription complex; nirmatrelvir target edge) and Acute Hepatitis C Virus Infection (HCV polyprotein → NS3/4A processing → RNA replication; glecaprevir target edge). Remaining highest-value work includes COVID-19 polymerase targets, chronic Hepatitis C NS5B/NS3/4A/NS5A nodes, Hepatitis B RT suppression gated by cccDNA, Acquired Immunodeficiency Syndrome as the flagship five-module conformer (RT + integrase + protease + entry + reservoir).
• [ ] Decide how to model Influenza baloxavir (cap-dependent endonuclease, PA subunit) vs oseltamivir (neuraminidase release): endonuclease is closer to a polymerase-complex target than to assembly/release — may warrant its own node rather than forcing it into viral_assembly_release_inhibition.
• [ ] Short survey of antiviral drug-mechanism KBs / resistance databases (DrugMechDB, Stanford HIVdb) — see §0; borrow vocabulary, sharpen the niche.
• [ ] Record the host-directed-vs-direct-acting module decision in the design register.

Note on Influenza/baloxavir and "endonuclease": influenza lacks an RdRp of the picornavirus type; its polymerase is a heterotrimer (PB1/PB2/PA), and baloxavir inhibits the PA cap-dependent endonuclease. Model it as a polymerase- complex subunit target, not a generic release inhibitor.