Delivery systems: recording what carries a drug

A treatment's carrier — the lipid nanoparticle, the liposome, the albumin particle, the viral vector — is recorded in a delivery_system block on the Treatment, alongside any targeting ligand on that carrier and what the carrier is aimed at.

treatments:
- name: Vutrisiran
  therapeutic_modality: SIRNA
  delivery_system:
    delivery_platform: CONJUGATE
    targeting_ligand: GALNAC
    targeting_receptor:
      preferred_term: ASGR1
      term:
        id: hgnc:742
        label: ASGR1
    target_cell_types:
    - preferred_term: hepatocyte
      term:
        id: CL:0000182
        label: hepatocyte

Why this is not inside `oligonucleotide_details`

It used to be. delivery_platform and conjugation were slots of OligonucleotideDetail, reachable only from a Treatment whose therapeutic_modality was ANTISENSE_OLIGONUCLEOTIDE or SIRNA. That put the carrier axis inside the payload chemistry, and the consequences were visible in the KB:

nab-sirolimus (Perivascular_Epithelioid_Cell_Neoplasm) is an FDA-approved albumin-bound nanoparticle. Its modality is SMALL_MOLECULE, so the carrier survived only in a free-text preferred_term, sirolimus albumin-bound nanoparticles.
Liposomal irinotecan in NALIRIFOX (Pancreatic_Ductal_Adenocarcinoma) is the one component of that regimen that distinguishes it from FOLFIRINOX, and that difference is entirely the carrier. Same problem.
MRNA_THERAPY has been a permissible therapeutic_modality with zero uses across the KB. It is a modality defined by its carrier, and there was no slot to name one.

None of those is an oligonucleotide, and a query for "every treatment in the KB delivered in a nanoparticle" could not have returned any of them.

The four attributes, and what each claims

Slot	Claim
`delivery_platform`	What carries the agent at all — unformulated, a ligand conjugate, an LNP, a liposome, an albumin particle, a viral vector
`targeting_ligand`	What is on the carrier (or on the agent) that drives uptake
`targeting_receptor`	The receptor or antigen that ligand binds, bindable to HGNC
`target_cell_types`	The cell type the carrier is aimed at, bindable to CL

delivery_platform and targeting_ligand are orthogonal, in both directions. Patisiran is UNCONJUGATED and LIPID_NANOPARTICLE; vutrisiran is GALNAC and CONJUGATE. A nanoparticle may also carry a ligand on its own surface — an antibody-coated mRNA-LNP is ANTIBODY and LIPID_NANOPARTICLE — which is the case that motivated generalizing the ligand slot beyond covalent attachment to an oligonucleotide.

LIPOSOME is not a spelling of LIPID_NANOPARTICLE. The ionizable lipid in an LNP is what destabilizes the endosomal membrane on acidification and releases a nucleic-acid payload; a PEGylated phospholipid bilayer vesicle carrying an already cell-permeant cytotoxic does no such thing. It changes biodistribution and toxicity instead. Collapsing them would erase the reason one exists.

An untargeted carrier is a normal record. A PEGylated liposome accumulates passively: it has no ligand, no receptor, and no target cell type. Leave those three slots absent rather than asserting a target the formulation does not have. Naming a targeting_receptor while setting targeting_ligand: UNCONJUGATED is a contradiction and is gated.

targeting_receptor is the receptor, not the ligand. A receptor may be reachable by more than one ligand, and the ligand's chemical class already has a slot.

Dosing interval lives on `Treatment`, not here

dosing_interval / dosing_interval_days apply to any treatment, so they stay where they were. They are worth reading next to a delivery_system, because the carrier is usually why the interval is what it is — vutrisiran's quarterly subcutaneous dose against patisiran's three-weekly infusion with premedication is a difference in carrier, not in mechanism or target.

Two homes, for now

oligonucleotide_details.delivery_platform and oligonucleotide_details.conjugation are still valid. They were kept rather than removed for the reason recorded in Retired Enum Values: retiring a spelling invalidates every PR already in flight that uses it, and ~44 oligonucleotide entries carry these slots. conjugation is marked deprecated in favour of targeting_ligand; both render.

So two spellings are legal, and just check-delivery-system is what stops them drifting:

just check-delivery-system                 # gate (runs in `just qc`)
just check-delivery-system --format list   # the full census, including the worklist
just check-delivery-system --strict        # also gate on the redundant case

It gates on three things, each a real defect:

CONFLICT — the same fact in both places with different values. One is wrong and nothing downstream can tell which; the renderer resolves delivery_system first, so the nested value is hidden rather than surfaced.
EMPTY — a delivery_system carrying no carrier fact. It renders as nothing; its absence says the same thing more honestly.
LIGANDLESS_TARGET — a receptor named alongside targeting_ligand: UNCONJUGATED.

It reports, and does not gate on:

DUPLICATE — the same fact in both places with the same value. Harmless today, one edit from a CONFLICT.
LEGACY — the carrier recorded only in the nested block. This is the migration worklist and currently holds 81 findings. Gating on it would turn every oligonucleotide entry in the KB red for a change none of their curators made.

New treatments — oligonucleotide or not — use delivery_system. Only ATTR_Amyloidosis's vutrisiran has been migrated, as the worked targeted example; the rest of the oligonucleotide corpus is deliberately left on the nested spelling.

Worked examples

Entry	Treatment	What it demonstrates
`ATTR_Amyloidosis`	Vutrisiran	All four slots: `CONJUGATE` + `GALNAC` + ASGR1 + hepatocyte, migrated off the nested spelling
`Pancreatic_Ductal_Adenocarcinoma`	NALIRIFOX	`LIPOSOME` on a regimen where only one component is carried, with evidence
`Perivascular_Epithelioid_Cell_Neoplasm`	Nab-Sirolimus	`PROTEIN_NANOPARTICLE`, and a `notes:` recording why a hypothesized uptake route is not recorded as targeting

INORGANIC_NANOPARTICLE has no worked example yet.

The case that motivated the change

An antibody-coated mRNA lipid nanoparticle is the shape that the old schema could not express at any point: the modality has no oligonucleotide_details block to hang a carrier on, the ligand is coupled to the particle rather than to the payload, and the whole therapeutic claim is which cell it reaches. Chen et al. (Sci Adv 2026, 10.1126/sciadv.aed9568) coat an mRNA-LNP with an anti-TREM2 antibody to reach tumor-associated macrophages rather than the tumor cells beside them, reporting a macrophage-to-tumour-cell uptake ratio of 5.1. In delivery_system that is:

  therapeutic_modality: MRNA_THERAPY
  delivery_system:
    delivery_platform: LIPID_NANOPARTICLE
    targeting_ligand: ANTIBODY
    targeting_receptor:
      preferred_term: TREM2
    target_cell_types:
    - preferred_term: tumor-associated macrophage

This snippet is illustrative only — the agent is preclinical and no KB entry curates it. It is here because it is the clearest statement of what the block is for: the mechanism claim of a targeted nanomedicine is a claim about the delivery system, not about the drug.