npbuilds Modality Lifecycle — The Live Maturity Map of Drug Delivery
Modalities, not just molecules, have life-cycles. A small molecule, a GalNAc-siRNA, and an in-vivo CRISPR editor are at radically different points on the same arc — and the single state variable that determines where a modality sits is not its biology but its delivery unlock: what is the delivery problem, and has the fix landed? Get this wrong and you either dismiss a class that just inflected (TPD before vepdegestrant) or over-credit one whose delivery wall is still standing (extrahepatic RNAi, in-vivo CAR in humans).
The core principle: the platform fix, not new biology, is what converts a stalled class. siRNA was dead-on-delivery for a decade until GalNAc conjugation gave it the liver; KRAS was "undruggable" until the 2013 covalent switch-II discovery; degraders waited on E3-ligase recruitment; ADCs were left for dead after gemtuzumab until site-specific linkers and DXd payloads. A high-profile failure followed by a modality re-engineering is a buy signal, not a kill signal. And the arc runs both ways: a modality can regress — AAV gene therapy slid backward in 2025 after Elevidys was tied to ≥3 deaths and an FDA dosing pause.
This skill is dual-use by design. For a clinician-scientist it is a map of where each modality actually stands and what unlocks the next tissue. For an investor it supplies the P(modality deliverable) term that the conviction scorer multiplies against P(biology holds), arc-position, and competitive timing — the discipline that stops a great target from being funded inside an undeliverable modality.
Maturity map: the per-modality S-curve placement, delivery-wall lookup, and platform-fix registry are versioned in
references/modality-maturity-map.md(a 2026 point-in-time snapshot — the most perishable data in the pillar).
How to Run
Input
| Parameter | Source | Required? |
|---|---|---|
| Modality (e.g., in-vivo CAR, GalNAc-siRNA, base editing, PROTAC) | User | Yes |
| Target tissue / compartment (liver, muscle, CNS, T-cell, tumor) | User | Recommended — delivery is tissue-specific |
| Specific asset / company (if scoring a candidate) | User | Optional |
| Indication | User | Optional (gates which delivery route matters) |
| Use mode: learn (field map) vs screen (conviction input) | User | Optional (default: screen) |
Steps
Step 1 — Locate the Modality on the Four-Stage Maturity Map
Assign one of four stages. The boundary tests are concrete, not vibes:
- Emerging — first-in-human to early Ph1; proof-of-mechanism not yet durable or safe at scale. (in-vivo CAR, prime editing, CAR-NK/allogeneic)
- Proving — Ph1/2 efficacy signals replicating; ≤1 approval; platform risk still live. (TPD, base editing, in-vivo CRISPR nuclease, TCR-T/TIL, oral biologics, saRNA)
- Established — multiple approvals; mechanism de-risked; scaling/access is the remaining work. (ADCs, ASO/siRNA, mRNA vaccines, ex-vivo CAR-T, ex-vivo CRISPR, RLT, lentiviral)
- Mature — dozens-to-hundreds of approvals; commoditized; innovation is incremental or in delivery. (small molecules, mAbs, peptides/macrocycles)
Start from the maturity-map table in references/quick-reference.md. Do not anchor on hype — anchor on approvals shipped × mechanism de-risking.
Step 2 — Identify the Binding Delivery Bottleneck
For the modality + target tissue, name the one delivery/enabling-technology constraint that gates the next stage. This is the load-bearing analytical step. The universal rate-limiter in 2026 is delivery, not biology:
DELIVERY-WALL LOOKUP
siRNA / ASO → extrahepatic reach (CNS, muscle, lung, immune cells)
in-vivo CRISPR/base → LNP tropism beyond liver
in-vivo CAR → T-cell- (or HSC/NK-) tropic LNP / retargeted vector
AAV → pre-existing immunity, no redosing, liver-detargeting capsids
mRNA therapeutics → repeat-dose tolerability, extrahepatic LNP
oral peptides/biologics→ gut proteolysis + epithelial permeability (often <2% F)
radioligand therapy → isotope supply chain (Ac-225) + radiopharmacy sites
cell therapies → "delivery" = manufacturing logistics + persistence/rejection
prime editing → editor payload size (hard to package)
Verify the current state of the wall against live data:
- ClinicalTrials.gov (c-trials MCP) — confirm phase/status/enrollment of the proof-point trials (e.g., HAELO for lonvo-z; CPTX2309 Ph1; RAG-17 NCT06556394 for SCAD/intrathecal siRNA).
- bioRxiv/medRxiv — delivery-platform preprints (tLNP, antibody-siRNA conjugates, novel capsids) lead approvals by 6-18 months; this is where the next unlock surfaces first.
- ChEMBL — bioactivity for the enabling chemistry (covalent KRAS warheads, E3-recruiting glues, payload IC50s).
- PubMed/PMC + Consensus — mechanism/review literature and evidence synthesis on whether the delivery claim actually replicates.
Step 3 — Determine Whether the Platform Fix Has Landed
Convert "delivery wall" into a binary-with-evidence: has the enabling-platform fix landed, partially landed, or not? The known fixes that convert a stalled class:
PLATFORM-FIX REGISTRY (fix, not biology, converts the class)
GalNAc conjugation → siRNA/ASO to liver (LANDED — inclisiran, vutrisiran)
covalent switch-II → KRAS druggability (LANDED — sotorasib 2021)
tri-complex / RAS(ON) → pan-RAS beyond G12C (LANDING — daraxonrasib Ph3/BTD)
E3-ligase recruitment → targeted degradation (LANDED — vepdegestrant 2025)
site-specific linker + DXd → ADC redemption (LANDED — T-DXd, Datroway, Emrelis)
T-cell-tropic LNP (tLNP) → in-vivo CAR (PARTIAL — CPTX2309 Ph1, NHP data only)
C16/antibody-siRNA conj. → extrahepatic RNAi (PARTIAL — muscle AOCs, SCAD preclinical)
low-seroprevalence capsid → AAV redosing (NOT LANDED — class regressed post-Elevidys)
multiplexed HLA editing → allogeneic persistence/rejection (NOT LANDED)
If the fix has LANDED → modality clears for that tissue. If PARTIAL → flag the specific trial whose readout closes it. If NOT LANDED → the modality is gated regardless of target merit.
Step 4 — Check for Regression and Maturity Signals
The arc is bidirectional and the late-arc tells matter:
- Regression flag — a marquee safety event reversing a class. AAV is the index case: Elevidys (DMD) tied to ≥3 liver-injury deaths and an FDA dosing pause in 2025, with Sarepta itself pivoting toward siRNA for redosing. A regressed modality borrows a worse transition matrix, not its pre-event one.
- Maturity / commoditization flag — convenience/cadence reformulation. When a class competes on route and dosing cadence rather than efficacy, the alpha has left: subcutaneous nivolumab/pembrolizumab (Keytruda Qlex, Sept 2025), oral GLP-1 orforglipron, oral PCSK9 enlicitide, twice-yearly inclisiran. Note this for the investor read — mature ≠ attractive entry.
Step 5 — Emit the P(modality deliverable) Term
Translate the position into a deliverability probability for the conviction scorer. Anchor to modality LOA base rates and apply delivery/CMC/novelty haircuts:
P(modality deliverable) = base_modality_LOA_proxy × delivery_state_multiplier × regression_penalty
base proxies (LOA, BIO/Informa class reads):
vaccines ~9.7% | biologics ~9.1% > small molecules ~5.7% | CGT ~5.3% bimodal
(CAR-T / AAV historically 13.6% in their validated indications)
delivery_state_multiplier:
LANDED for this tissue = 1.0
PARTIAL (proof in NHP/Ph1) = 0.6-0.8
NOT LANDED (wall standing) = 0.3-0.5
zero-clinical-history class = borrow nearest validated class matrix, never de novo optimism
regression_penalty:
active safety reversal = ×0.5-0.7 (AAV post-Elevidys)
none = ×1.0
For a zero-clinical-history modality, borrow the nearest validated class's transition matrix and apply explicit delivery/CMC/regulatory-novelty haircuts — never an optimistic de novo estimate. Platform risk concentrates in Phase 1→2.
Output
MODALITY LIFECYCLE READ — [Modality] × [Target tissue]
Date: [assessment date] | Mode: [learn / screen]
Maturity stage: [Emerging / Proving / Established / Mature]
Representative drugs: [examples]
Lead players: [companies]
Binding delivery bottleneck: [the one constraint gating next stage]
Current state: [wall standing / partially breached / breached]
Proof-point trial(s): [NCT# / readout + date] Source: [ClinicalTrials.gov / preprint]
Platform-fix status: [LANDED / PARTIAL / NOT LANDED]
The fix: [GalNAc / tLNP / E3-recruitment / capsid / etc.]
What closes it: [specific pending readout if PARTIAL]
Arc-direction flags:
Regression: [none / active — describe safety event]
Maturity/commoditization: [none / reformulation underway — route+cadence competition]
P(modality deliverable): [0.00-1.00]
= base [%] × delivery_multiplier [x] × regression_penalty [x]
Borrowed matrix (if zero-history): [nearest validated class + haircuts applied]
Bottom line:
Clinician read: [where it stands, what unlocks the next tissue]
Investor read: [deliverable enough to fund? entry attractive or commoditized?]
Error Handling
| Scenario | Response |
|---|---|
| Modality not on the map | Place it by analogy to the nearest mapped class; flag as zero-clinical-history and borrow that class's matrix with novelty haircuts |
| Conflicting stage signals (approval but class regressing) | Maturity stage = approvals shipped; apply regression_penalty separately. Record both — an Established class can still be a bad entry (AAV) |
| Delivery claim only in preprint / NHP | Mark PARTIAL; do not credit as LANDED until a human Ph1 dose-response confirms; name the trial that would close it |
| "AI-discovered" / novel-platform framing | Down-weight unless the platform has independent wet-lab + clinical validation; AI compresses discovery, not Phase 2/3 attrition |
| Tissue not specified | Default to the validated tissue (usually liver for oligo/LNP); flag that extrahepatic asks carry a separate, harder delivery state |
| Single company / single asset = the whole class | Treat as Emerging regardless of hype; one Ph1 asset is not a de-risked modality |
Cross-Domain Connections
- manufacturing-ip/modality-manufacturing (depends_on): The CMC/COGS counterpart — where this skill asks "can it reach the tissue?", that skill asks "can it be made at scale and cost?" Autologous CAR-T COGS ~$100-220K/dose and ex-vivo CRISPR's ~$2.2M price are manufacturing walls that cap an otherwise-deliverable modality. The two terms multiply.
- mechanism-risk-adjuster (depends_on): Supplies the modality-deliverability haircut that adjusts a target's base PoS. A genetically-clean target inside a NOT-LANDED modality must be discounted for platform risk, not credited at face value.
- frontier-discovery/frontier-conviction-scorer: Primary consumer — this skill emits the P(modality deliverable) term in the decomposition P(biology holds) × P(modality deliverable) × arc-position × competitive-timing.
- frontier-discovery/moa-analog-engine: Complementary — moa-analog-engine places the mechanism on its historical arc; this skill places the modality. A candidate needs both reads (KRAS the mechanism × small-molecule the modality).
- probability-of-success/pos-base-rates: Shares the modality LOA base rates; this skill is where the qualitative delivery state turns those rates into a program-specific deliverability probability.
- Dual use: A clinician-scientist runs this in learn mode to understand what gates each modality's next tissue; an investor runs it in screen mode to gate early opportunities before a clinical asset exists, then hands off to phase-weighted rNPV once one does.