designed biology

Drugs and gene edits designed by computation, and how far they have got in people.

Most drug candidates fail in human trials, and each takes a decade or more to reach patients. Software can now propose molecules and proteins, and gene editors can correct a faulty letter of DNA inside the body. Both promise faster, more precise medicines. The open question is whether designs made this way succeed in people more often than those made the old way.

Design is no longer the bottleneck.
In one 2024 analysis, AI-derived molecules passed Phase 1 safety trials at 80 to 90% (21 of 24 in its main sample); BIO puts the industry rate for 2011 to 2020 at 52%. That shows the methods make well-behaved molecules. It does not yet show they make better medicines.
Efficacy evidence is thin but arriving.
In the same analysis, 4 of 10 AI-derived molecules passed Phase 2, which the authors call comparable to historical averages; BIO puts the industry Phase 2 rate at 29%. Three computationally designed programs are now in Phase 3 (rentosertib, GB-0895 and zovegalisib), and a fourth, zasocitinib, has an FDA decision due in early 2027.
Gene editing has one approval.
An ex vivo CRISPR therapy has been on the market since December 2023. The first in vivo editor to be filed has an FDA decision date of 10 March 2027. Safety events paused other programs in 2025 before holds were lifted in 2026, so delivery and safety, not editing chemistry, appear to set the pace.
Regulators are adapting to the technology.
FDA draft guidance from October 2026 would let one license cover many patient-specific editors for the same disease. If adopted, it could lower the cost of treating rare mutations, with the clinical evidence needed decided case by case.

what would change the view

  • AI-derived Phase 2 pass rates fall to or below the 29% baseline as the number of readouts grows.
  • Two or more of the late-stage computationally designed programs (rentosertib, GB-0895, zovegalisib, zasocitinib) fail Phase 3 or receive a complete response letter.
  • A new in vivo editing safety event leads to class-wide clinical holds, or lonvo-z is not approved.

map

scoreboard How many AI-designed drugs have cleared each stage of human testing, and whether they clear it more often than drugs historically have. Counts of AI-designed drugs in the clinic mostly measure entries into Phase 1, which tests safety, not whether a drug works. Phase 2 is where most drugs fail and where the claim is tested.

Each company at its most advanced disclosed program, by approach. Filled dots: listed company or unit of one; hollow: private. The rule marks Phase 2, where the claim is tested. Highlighted: Casgevy (approved December 2023), zasocitinib (FDA decision due 1Q27), lonvo-z (FDA decision date 10 March 2027) and rentosertib (Phase 3 since September 2026). Structure- and physics-based design is shown as its own lane because trackers differ on whether it counts as AI-derived.

Share of drugs passing each phase: industry 2011 to 2020 (BIO, all modalities) against AI-derived molecules in one 2024 analysis (21 of 24 in Phase 1, 4 of 10 in Phase 2; the whisker shows the reported 80 to 90% range). The authors compare their figures with somewhat different industry ranges. No AI-derived drug has completed Phase 3, so that point is empty. Of 117 AI-derived assets that entered trials by December 2025, 60 had completed Phase 1, 8 Phase 2 and none Phase 3.

players

Listed companies with a design or editing platform and a program in human testing, or an approved product. Private companies are included only if a candidate has been dosed in a registered human trial. On that test, Isomorphic Labs, Xaira and Chai Discovery are not yet included. Funding amounts are not shown for anyone.

Insilico Medicinelisted, HKEX: 3696
AI small molecule (target and molecule) · Phase 3, rentosertib (first patient dosed Sep 2026). Listed in Hong Kong in December 2025; a second program, garutadustat, is in Phase 2a.
Recursionlisted, Nasdaq: RXRX
AI small molecule (phenomics) · Phase 2, REC-4881 (Oct 2026). Registrational study targeted for 2H27.
Generate Biomedicineslisted, Nasdaq: GENB
generative protein and antibody · Phase 3, GB-0895 (SOLAIRIA-1 and -2). Long-acting antibody dosed every six months in severe asthma.
Abscilisted, Nasdaq: ABSI
generative antibody · Phase 1/2a, ABS-201 (Aug 2026). Interim proof-of-concept data in hair loss guided for 2H26.
Iambic Therapeuticslisted, Nasdaq: IAM
AI small molecule · Phase 1/1b, IAM1363 (Oct 2026). Listed in October 2026; HER2-targeted program.
Schrödingerlisted, Nasdaq: SDGR
physics-based design · Phase 1, SGR-1505 (May 2026). Design partner on molecules now developed by other companies, including zasocitinib.
Takedalisted, NYSE: TAK
structure-based design (from Nimbus) · Filed, zasocitinib (Sep 2026). FDA decision targeted 1Q27 in plaque psoriasis.
Relay Therapeuticslisted, Nasdaq: RLAY
computational design · Phase 3, zovegalisib (ReDiscover-2). Enrollment update guided by year-end 2026.
Vertexlisted, Nasdaq: VRTX
ex vivo editing · Approved, Casgevy (Dec 2023). Commercial partner for Casgevy, with CRISPR Therapeutics.
CRISPR Therapeuticslisted, Nasdaq: CRSP
CRISPR nuclease, ex vivo and in vivo · Approved, Casgevy; Phase 1, CTX310. CTX310 one-year data published in August 2026.
Intellia Therapeuticslisted, Nasdaq: NTLA
in vivo CRISPR nuclease · Filed, lonvo-z (FDA decision date 10 Mar 2027). Clinical holds on its nex-z Phase 3 trials were lifted in 2026.
Beam Therapeuticslisted, Nasdaq: BEAM
base editing · Pivotal cohort, BEAM-302 (first dosed Jul 2026). Pivotal cohort sits inside a Phase 1/2 study, designed for accelerated approval.
Prime Medicinelisted, Nasdaq: PRME
prime editing · Phase 1/2, PM577a (first patient Oct 2026). Wilson disease; initial data expected 2027.
Eli Lilly (Verve)unit of Eli Lilly, NYSE: LLY
in vivo base editing · Phase 1b, VERVE-102 (May 2026). LDL cholesterol reduced by up to 62% after one dose.
YolTech Therapeuticsprivate
in vivo base editing · First-in-human study, YOLT-202 (Feb 2026). Shanghai-based; FDA cleared a Phase 2/3 study of YOLT-202 in March 2026, after early data from an investigator-initiated trial in China.
Arbor Biotechnologiesprivate
in vivo CRISPR nuclease · Phase 1/2, ABO-101 (registered trial). Primary hyperoxaluria type 1.

resources

CRISPR Medicine News
Independent gene-editing news with a quarterly overview of editing trials worldwide.
Fierce Biotech
Daily news on clinical readouts, regulatory decisions and company announcements.
BiopharmaTrend
Analysis and news on AI in drug discovery and new biotech tools.
BIO, Clinical Development Success Rates 2011-2020
Phase-by-phase success rates across the industry; the baseline used on this page.

notes

Positions are approximate and reflect editorial judgment, as of 3 November 2026; each company appears at its most advanced disclosed program in each approach. Phase 1/2 trials are placed at Phase 1 unless a Phase 2 portion is reported as dosing; Beam's pivotal cohort is placed at Phase 3 because the company describes it as pivotal. "AI-derived" has no regulatory definition and trackers disagree on what counts. The AI pass rates come from one 2024 analysis with small samples, and the industry baseline is not matched by disease area, so the comparison is indicative. Sequencing and synthesis tools are outside this page for now. Not investment advice.