Skip to content

Explore-mode strategic report — RNAi & oligonucleotide therapeutics for obesity (from one sentence)

Thesis (the only input): "siRNA, RNAi and antimiR oligonucleotide therapeutics for obesity and metabolic disease, including INHBE/activin-E and ALK7 (ACVR1C) programs, adipocyte, appetite and hepatic metabolic targets, and preclinical China- and Korea-based candidates." Date: 2026-07-09 · Mode: Explore (cold-live discovery, scored vs the obesity-RNAi benchmark deck)

The asset list is discovered from the thesis, not seeded: the system queried live public sources cold (no company names given), served patents from a deck-agnostic concept corpus at $0/query, and scored what it found with the same matchers the closed-world test uses. The shareable analyst-facing version (no methodology/benchmark) is obesity_rnai_explore_SAMPLE_20260709.md. Two honest differences from the siRNA cardiometabolic report: this deck has no deal or insights/views ground truth (assets + companies only), and this pass was discovery + recall-focused — the strategic read (§6) is analyst synthesis over the discovered corpus, not a machine-generated multi-KIQ briefing.

1. Bottom line — recall

Cold, from one sentence, scored against the hand-built deck (20 assets · 15 companies · tiers 14 core / 5 metabolic-adjacent / 1 questionable), using the same matchers the closed-world test uses:

metric recovered of deck note
Companies 11 15 (73%) up from 39% before the concept-corpus rebuild
Assets (by sponsor) 16 20 (80%) ~matches the 85% closed-world Monitor ceiling
Assets (by code) 5 20 (25%) codes rarely appear in patent text
Deals 0 this deck tracks assets + companies only (no deal axis)
  • The concept-corpus rebuild is what did it. Recall lifted from 39%/55% (companies/assets) to 73%/80% purely by rebuilding the patent corpus broad + deck-agnostic — gpatents first-surfaced 10 of the 11 recalled companies (§5).
  • Beat the earlier ceiling call: BaseCure and OliX were predicted unreachable but surfaced anyway (§3).
  • Cost: ~$0.01 discovery (metered LLM tokens; deterministic scope, gpatents-db avoids BigQuery).
  • Precision is 1% by design — the product is recall + flagged leads, not a vetted shortlist. The queued fix (an evidence-gated corpus-gate, 900→134 targets, recall-safe) is a follow-up PR — §7.

Scorecard — Explore vs the closed-world Monitor pipeline

Unlike the siRNA deck, this deck carries no deals and no insights/structured-views ground truth, so the scorecard is recall-focused (there is no I01–I09 / V01–V07 axis here).

metric Explore (cold, this run) Monitor (closed-world, seeded)
Companies 11/15 (73%) 11/11 core (100%)
Assets (by sponsor) 16/20 (80%) 12/14 core (85%)

Live Explore now ~matches closed-world Monitor from a cold thesis — where the gap used to be 55%/39%.

Per-asset depth — identity recovered, depth not (flagged)

Over the 16 discovered assets, Ogur recovers identity but not depth:

field recovered how
Company / Target / Modality 16/16 (target coarse) discovery sponsor-attribution; company-level target set
Phase / stage 0/16 (coarse) patent-stage records carry no structured trial phase
Differentiation / MoA 0/16 structural gap — lives in the patent BODY, not fetched this run

2. Thesis → scope (the Explore front door)

The one sentence decomposes to a structured ScopeSpec, then expands via the Open Targets disease→target graph:

  • Modalities: siRNA, RNAi, antimiR, oligonucleotide
  • Targets (HGNC): INHBE (anchor), ACVR1C/ALK7, GRB14, ANGPTL¾/8, APOC3, AGT, FAM13A, GDF15, PNPLA3, HSD17B13, LEPR, GIPR, … (LLM thesis targets ∪ 983 OT-expanded genes → 1,008 total)
  • Indications: obesity, T2D, metabolic syndrome, MASH/NASH, NAFLD, dyslipidemia, hypertriglyceridemia

The ~1,008-target expansion is deliberately broad for recall — and is the source of the low precision (§7). Every deck-hit target survives the (queued) evidence-gate, so the gate reclaims precision without costing recall.

3. Discovered assets — with provenance (16 of 20)

Each recovered asset with the source that first surfaced it:

asset sponsor target stage (public) surfacing source
ARO-INHBE Arrowhead Pharmaceuticals INHBE Phase ½ gpatents · sec
ARO-ALK7 Arrowhead Pharmaceuticals ACVR1C (ALK7) preclinical gpatents
WVE-007 Wave Life Sciences INHBE Phase 1 openalex
ALN-4324 Alnylam GRB14 preclinical gpatents
ALN-2232 Alnylam ACVR1C preclinical gpatents
(INHBE) Alnylam INHBE patent-stage gpatents
(ANGPTL4) Alnylam ANGPTL4 patent-stage gpatents
BC-006 BaseCure Therapeutics INHBE Phase 1 gpatents · openalex
Cotsiranib Sirnaomics TGF-β1 + COX-2 patent-stage gpatents
(INHBE) Suzhou Siran Biotechnology (SiranBio) INHBE patent-stage gpatents
(ACVR1C) Suzhou Siran Biotechnology (SiranBio) ACVR1C patent-stage gpatents
(ANGPTL3) Shenzhen Salubris ANGPTL3 patent-stage gpatents
(INHBE) Shanghai Argo Biopharma INHBE patent-stage gpatents
OLX702C OliX Pharmaceuticals APOC3 patent-stage gpatents
(FAM13A) Amgen FAM13A patent-stage gpatents
(AGT) Sirius Therapeutics AGT patent-stage gpatents

Three recoveries that beat the earlier ceiling call. BaseCure (INHBE) and OliX (OLX702C / APOC3) were predicted unreachable ("ISRCTN-only" / "Korean-DART-only") — the broad concept scan found their patents anyway. Sirius (AGT) surfaced via a patent drafted as "double-stranded nucleic acid" that the narrower modality gate had dropped. All 16 are two-factor grounded (sponsor + on-thesis target/modality), one click from the source record.

4. Missed assets — why each is unreachable (4 of 20)

All four misses are corpus-ceiling: no reachable public footprint under an obesity-RNAi thesis, not a search failure. They match the closed-world's documented unreachables.

asset company target why unreachable
CCT-217 Canary Cure Therapeutics ZNF423 + CNR1 private US company, no public filing of any kind
BF-114 BullFrog AI SPTBN1 target off the obesity axis; code lives only in the SEC filing body (full-text axis, not run live)
RES-020 Resalis Therapeutics miR-22 microRNA-only program; thin/absent patent footprint
Hadasit CD59 off-thesis (complement biology), not an obesity mechanism — correctly not chased

The code-level miss rate (5/20) is the same disclosure limit: asset codes (ARO-, ALN-, OLX-) rarely appear in patent title/abstract text, so a patent-sourced recovery names the company + target but not always the code.

5. Per-source contribution (which source first surfaced each recalled company)

source deck companies first-surfaced
gpatents (rebuilt concept corpus) 10
openalex 1

The rebuilt concept-patent corpus is the workhorse — 10 of 11 recalled companies, including every China patent-stage program (Salubris, Argo, SiranBio) and the three ceiling-beaters (Sirius, BaseCure, OliX). OpenAlex first-surfaced Wave (WVE-007) via a research affiliation. CT.gov / SEC / CDE corroborated but rarely first-surfaced a company — the concept corpus reaches the patent-stage innovators the US trial registry never sees.

6. Net-new leads (top core, of 49) — flagged, need primary-source verification

Non-deck players the scope-fit classifier labelled core. The standout finding: the GLP-1 incumbents and the RNAi platforms are both moving into obesity RNAi.

company modality latest activity
Novo Nordisk siRNA 2026-02
Eli Lilly siRNA 2026-05
Novartis RNAi 2026-05
Regeneron siRNA 2026-04
Silence Therapeutics siRNA 2026-06
Ionis siRNA 2026-03
Boehringer Ingelheim RNAi 2025-07
Suzhou Ribo Life Science siRNA 2026-04
Insitro RNAi (GCK) 2026-02
Empirico siRNA (ANGPTL) 2026-04
Eddingpharm siRNA (EDP167) 2026-03
Frontier Biotechnologies siRNA 2026-02
Qilu Pharmaceutical RNAi 2026-03
BeBetter Med siRNA 2026-02

(49 core + 846 honorable-mention in the artifacts; academic labs and patent-inventor person-names filtered. Target sets are company-level, not per-asset — the disclosure limit of cold discovery.)

7. Strategic read — synthesized from the discovered landscape

Grounded in the cited assets, companies and filings above. This pass was discovery + recall-focused; the analysis below is analyst synthesis over the discovered corpus — the automated multi-KIQ synthesizer briefing (as run for the siRNA landscape) is the natural follow-up.

7.1 Executive summary

The obesity-RNAi landscape has converged on a single anchor target — INHBE (inhibin βE / activin-E) — the clearest mechanism cluster outside the GLP-1 axis: Arrowhead (ARO-INHBE, Phase ½), Wave (WVE-007), Alnylam, BaseCure (BC-006) and the China cluster (SiranBio, Shanghai Argo) all file on INHBE, extending to its receptor ALK7 (ACVR1C) (Arrowhead ARO-ALK7, Alnylam ALN-2232, SiranBio). The thesis is a durability and mechanism play against GLP-1 incumbency — a GalNAc-siRNA silencing a liver-secreted obesity hormone offers quarterly-to-biannual dosing and a fat-specific, genetically-validated mechanism vs weekly injectable GLP-1s with tolerability and lean-mass concerns. The single most important discovery is that the GLP-1 incumbents themselves — Novo Nordisk and Eli Lilly — are filing double-stranded-oligo IP, alongside Novartis, Regeneron and Boehringer Ingelheim: the market leaders are hedging into the modality that could unseat their own franchises. The field is early (Phase 1 / preclinical) and heavily China-weighted, so the in-license window for differentiated, non-INHBE targets is open now and narrowing.

7.2 Strategic implications

INHBE is crowded; the edge is the second target. With six+ recovered players on INHBE, the anchor is table stakes. The white space is the diversifier axis the discovery surfaced thinly — GRB14 (Alnylam ALN-4324), FAM13A (Amgen), ANGPTL4 (Alnylam), and the appetite/adipose GWAS genes. A program leading with a validated non-INHBE target avoids a six-way race.

The incumbents' RNAi filings are the signal to watch. Novo Nordisk and Eli Lilly filing siRNA IP in obesity is the market defining its own next wave. Internal build (vs visible licensing) suggests the near-term BD action is acquiring the platform/target, not the asset — the Chinese INHBE innovators (SiranBio, Argo, Salubris) and the receptor-axis players are the likeliest targets before incumbent programs mature.

China owns the early-innovation density. A majority of recovered non-US assets are Chinese, several patent-stage only (Salubris ANGPTL3, Argo/SiranBio INHBE, Sirius AGT). Patent-only visibility means they are reachable now, before a trial registration raises their profile and price.

7.3 What to watch (forward calls)

  • Arrowhead's ARO-INHBE (Phase ½) is the field's clinical bar — its first weight-loss + body- composition readout sets the durability/efficacy benchmark every INHBE follower is measured against.
  • A GLP-1 incumbent (Novo or Lilly) discloses or in-licenses an obesity-RNAi program within 12–18 months — the internal IP filings surfaced here are the leading indicator.
  • ALK7/ACVR1C is the next consensus target after INHBE — Arrowhead, Alnylam and SiranBio all filing there.

7.4 Key intelligence questions

Is there a consensus anchor target? (high) — Yes: INHBE/activin-E, with ALK7 (ACVR1C) as the paired receptor mechanism. The durability/mechanism alternative to GLP-1.

Are the GLP-1 incumbents entering the modality? (high; leads flagged) — Novo Nordisk and Eli Lilly both surfaced as core net-new leads on double-stranded-oligonucleotide IP, with Novartis, Regeneron and Boehringer. Company-level patent evidence; specific obesity assets need primary-source verification.

Does China dominate early innovation? (high) — A majority of recovered non-US assets are Chinese (Salubris, Argo, SiranBio, Sirnaomics, BeBetter, Suzhou Ribo, Qilu), most patent-stage.

How clinically mature is the field? (medium) — Overwhelmingly early: front-runners are ARO-INHBE (Ph½), WVE-007 (Ph1), BC-006 (Ph1); the rest are preclinical / patent-stage with no structured phase.

Where is the BD white space? (medium) — Off the crowded INHBE anchor: GRB14, FAM13A, ANGPTL4 and the appetite/adipose GWAS axis.

8. Run details, reproducibility & cost

  • Discovery: cold-live (--corpus live, no seeded DB, no company-derivation), --patents-backend gpatents-db (deck-agnostic concept corpus, $0 BigQuery), OT bio-target expansion on. 4 rounds (round 4 = +0 novel → converged). 140 queries · 507 source API calls · ~$0.01 LLM · 630s wall-clock. Discovered 1,941 companies + 30 asset codes from 8,408 backing signals.
  • Concept corpus: archived_data/obesity_concept_gpatents_v2.jsonl9,443 rows / 2,198 distinct assignees, sha256 7dc9bd67a312fb85… — built target×modality (no company names) via scripts/seed/seed_gpatents_concept_obesity.py (one ~$1.40 BigQuery scan, priced free via dry-run first). The 16 MB corpus is a local artifact and is not committed; its builder is — so this run is re-runnable rather than byte-reproducible. Checksums, the exact regeneration command and the --extra gpatents silent-zero trap are recorded in archived_data/obesity_rnai_explore/FROZEN_INPUTS.md §2 (in the repository, not on the docs site). Broadening the target net (908 genes incl. the previously-empty GRB14/PNPLA3) and the modality vocabulary (Sirius's "double-stranded nucleic acid" idiom) is what lifted recall from 39%/55% to 73%/80%.
  • Precision caveat (by design): company precision 1% — the ~1,008-target OT expansion drags in off-thesis siRNA (cosmetic / agricultural). The queued fix is an evidence-gated corpus-gate between OT expansion and the fan-out (keep only OT genes a siRNA patent in the concept corpus names): 900 → 134 on this corpus, recall-safe (every deck-hit gene survives). Its own follow-up PR — see ~/.claude/plans/gate-pr-next-session.md.
  • Honesty line: the concept corpus is concept-seeded (targets × modality), never company-seeded — assignees are discovered, not planted. Targets/indications are analyst biology + OT genetics (deck-agnostic); the deck's company/asset columns were used only for recall scoring.
  • DB safety: the live ogur.db was untouched; all discovery went through the local concept corpus + cached HTTP sources.

Open follow-ups

  • Implement the evidence-gated corpus-gate (reclaim precision without losing recall) — its own PR.
  • Run the automated multi-KIQ synthesizer over the discovered corpus (the §7 briefing, machine-generated).
  • Per-asset depth (mechanism / MoA / milestone) via patent full-text fetch — the fetch-depth roadmap item.

Artifacts: archived_data/obesity_live_run5_recall/ (companies + assets CSV, gap analysis) · concept corpus archived_data/obesity_concept_gpatents_v2.jsonl · shareable sample docs/obesity_rnai_explore_SAMPLE_20260709.md · branch eval/obesity-rnai-setup.