How AI Designs a Drug: From Target to Clinical Trial
AI helps choose a disease target and generate candidate molecules, but lab tests and Phase 1 to 3 trials still decide what works. One candidate credited to generative AI entered Phase 3 in 2026.
// ai in biotech
A dated, sourced log of milestones since 2021 for drug candidates that their developers say were found or designed with AI: trial starts, results, regulatory steps, discontinuations and selected deals.
Researched and fact-checked by AI, with no human review. How we verify
26 entries from 29 sources. Checked for news weekly; last checked
This tracker logs dated milestones from April 2021 to October 2026, plus events scheduled into early 2027. It covers drug candidates whose developers credit artificial intelligence (AI) or machine learning. They say it helped pick the biological target, design the molecule or match a compound to a disease. The earliest entry is an A2a receptor blocker entering human trials in April 2021.
An event gets an entry if it is a trial start, a published or company-reported result, or a regulatory designation or filing. A discontinuation or a selected partnership or acquisition deal also gets one. Phases follow the FDA's definitions. Phase 1 studies safety and dosage. Phase 2 studies efficacy and side effects. Phase 3 studies efficacy and adverse reactions in 300 to 3,000 volunteers. Company results and targets are attributed to the company, and future dates are marked as scheduled.
None of the sources cited here reports a regulatory approval for any listed drug as of October 6, 2026. The furthest along, Takeda's zasocitinib, was accepted for FDA review in September 2026. Schrödinger says it co-invented that molecule with Nimbus Therapeutics using a platform it describes as physics plus AI. Takeda's release on the filing does not credit AI. The label "AI-discovered" therefore rests on developers' own descriptions.
Left out on purpose: private funding rounds, preclinical candidates, software and model releases, and market forecasts.
Targets set by the organizations named. Dates like these often move.
Scheduled
The FDA's target action date for Takeda's psoriasis pill falls in the first quarter of 2027, HCPLive reported on September 14, 2026, citing Takeda. The date was set under the Prescription Drug User Fee Act when the agency accepted the application. A target date is not an approval.
Source: FDA Priority Review Granted to Zasocitinib for Moderate to Severe Psoriasis, HCPLive
Scheduled
Chief executive Demis Hassabis said the Google-backed company expects its first clinical trials by the end of 2026, Reuters reported. He spoke at a World Economic Forum event in Davos on January 20, 2026. A year earlier he had targeted the end of 2025. This is a company goal, not a result.
Source: Google-backed Isomorphic Labs delays clinical trial timeline, Reuters (via Investing.com)
Scheduled
Recursion said on August 5, 2026, that updated TUPELO trial data would be presented on November 2 at the CGA-IGC annual meeting. The trial is in familial adenomatous polyposis (FAP), a hereditary condition marked by intestinal polyps. Recursion said FDA talks on a registrational path began in the first half of 2026.
Takeda said that in its Phase 3 LATITUDE Atlas study, 36.5% of 301 patients on zasocitinib had completely clear skin (PASI 100) at week 16. It said the figure was 13.9% of 303 on the comparator deucravacitinib. The company-reported data were presented at the EADV Congress 2026, after topline results in June.
Takeda said the FDA accepted its New Drug Application for zasocitinib in moderate-to-severe plaque psoriasis. The drug is a once-daily pill that blocks the enzyme TYK2. Takeda cited results from nearly 3,000 patients. It said the European Medicines Agency also accepted a filing and no regulator has approved the drug.
Insilico Medicine said the first patient was dosed at Peking Union Medical College Hospital in GENESIS-IPF-3. It said the 52-week placebo-controlled trial is expected to enroll 320 people with idiopathic pulmonary fibrosis at 47 centers in China. Insilico calls it the first Phase 3 trial of a generative AI drug.
Generate's IPO prospectus says the first patient in SOLAIRIA-1 was dosed on January 26, 2026. It says SOLAIRIA-1 is one of two Phase 3 trials in severe asthma begun in December 2025. It calls GB-0895 an antibody against the protein TSLP, computationally engineered with its Generate Platform for dosing every six months.
Isomorphic Labs announced a multi-target research collaboration with Johnson & Johnson. It said its AI drug design engine can generate candidates across several drug types, including small molecules, antibodies, peptides and molecular glues. Isomorphic's announcement page listed no financial terms.
Source: Isomorphic Labs Enters into a Research Collaboration with Johnson & Johnson, Isomorphic Labs
Schrödinger said SGR-1505, its MALT1 inhibitor, received FDA Fast Track and Orphan Drug designations in 2025 for Waldenström macroglobulinemia, a B-cell cancer. Its 2026 priorities include finishing Phase 1 data packages and exploring partnerships for SGR-1505 and SGR-3515. Neither is an approval.
Takeda said two Phase 3 trials, with 693 and 1,108 participants, met both main goals against placebo at week 16. It said they also met all 44 ranked secondary endpoints, which included comparisons with the active comparator apremilast. These were topline, company-reported results for an unapproved drug.
Recursion reported a median 43% drop in polyp burden after 12 weeks of REC-4881 (12 evaluable patients). It reported a median 53% drop at week 25, 12 weeks off treatment (11 patients). The data come from the single-arm Phase 2 part of TUPELO. Recursion says AI-driven cell screening linked the MEK1/2 inhibitor, licensed from Takeda, to FAP.
Schrödinger ended work on SGR-2921, a CDC7 inhibitor. The drug was in a Phase 1 dose-escalation study in relapsed or refractory acute myeloid leukemia (AML) or high-risk myelodysplastic syndromes. Schrödinger said the drug was considered to have contributed to two AML patients' deaths, despite early signs of activity.
Source: Schrödinger Announces Discontinuation of SGR-2921 Program, Schrödinger
A 12-week placebo-controlled trial randomized 71 people with idiopathic pulmonary fibrosis at 21 Chinese sites; safety was the primary endpoint. Mean forced vital capacity, a lung function measure, rose 98.4 mL on 60 mg once daily and fell 20.3 mL on placebo. The authors urged larger, longer trials.
Recursion ended REC-994 in cerebral cavernous malformation after long-term data showed no promising MRI or functional trends. It ended REC-2282 in an indication it labels NF2, citing limited tumor shrinkage. It said it would consider out-licensing REC-3964. It also paused REC-4539.
Exscientia became a wholly owned subsidiary of Recursion, joining two AI drug discovery companies. Recursion said the combined pipeline held more than 10 clinical and preclinical programs. It said the combined company had received about $450 million in partnership payments to date.
Recursion said its Phase 2 SYCAMORE trial of REC-994 met its main goal, safety and tolerability. It said an early version of its platform linked REC-994 to cerebral cavernous malformation. MRI trended toward smaller lesions at 400 mg; patient and physician-reported outcomes had not improved by 12 months.
The peer-reviewed paper described using AI to identify the protein TNIK as a fibrosis target and to design the inhibitor INS018_055, later named rentosertib. It said this took roughly 18 months from target discovery to preclinical candidate. It reported a Phase 1 trial in 78 healthy participants in New Zealand.
Isomorphic, launched in 2021 as a spinout of Google DeepMind, announced two partnerships, TechCrunch reported. It reported $45 million upfront from Eli Lilly, with up to $1.7 billion in milestones. It reported $37.5 million upfront from Novartis, with up to $1.2 billion. Milestone sums are maximums and exclude royalties.
Source: Isomorphic inks deals with Eli Lilly and Novartis for drug discovery, TechCrunch
Exscientia said it was discontinuing internal development of EXS21546, which it described as an A2A receptor antagonist. It said it was closing the Phase 1/2 IGNITE trial in relapsed or refractory kidney and lung cancers. It framed the move as pipeline prioritisation in favor of its CDK7 and LSD1 inhibitors.
Source: Exscientia plc Form 6-K, Report of Foreign Private Issuer (November 2023), Exscientia (SEC filing)
Insilico said the first patients had been dosed in a Phase 2 trial of INS018_055 for idiopathic pulmonary fibrosis. It described a randomized, double-blind, placebo-controlled study of 12 weeks of oral treatment. It planned to recruit 60 patients at about 40 sites in the United States and China.
Insilico said the FDA had granted Orphan Drug Designation to INS018_055 for idiopathic pulmonary fibrosis. Its release lists incentives such as tax credits and user fee exemptions; the designation is not an approval. Insilico said Phase 1 topline data indicated favorable safety and tolerability.
Takeda announced a deal to acquire NDI-034858, the drug later named zasocitinib, from Nimbus Therapeutics. It agreed to pay $4 billion upfront plus two $1 billion sales milestones. The deal followed positive Phase 2b psoriasis results. The release describes Nimbus as pairing computational methods with machine learning.
Sanofi agreed to pay Exscientia $100 million upfront. Milestone payments could total up to about $5.2 billion if all milestones for all programs are achieved. The deal is to develop up to 15 small-molecule candidates in oncology and immunology using Exscientia's AI-driven platform.
Recursion announced a collaboration in neuroscience and one oncology indication with a $150 million upfront payment. Roche and Genentech may start up to 40 programs, each with potential milestones above $300 million. Recursion said its neural networks would analyze cell data to find new biology.
Insilico Medicine said the first healthy volunteers had been dosed in Australia in a first-in-human microdose trial of ISM001-055. It said the fibrosis drug candidate, later named rentosertib, was given intravenously. The company said AI was used both to find the drug's target and to design the molecule.
Source: AI-discovered Novel Antifibrotic Drug Goes First-in-Human, Insilico Medicine
Evotec said an A2a receptor antagonist it co-invented with Exscientia had entered human trials for adults with advanced solid tumors. It said the drug was co-invented using Exscientia's Centaur Chemist AI design platform. Exscientia CEO Andrew Hopkins said the candidate was found within eight months of the project's start.
Source: Evotec and Exscientia announce start of human clinical trials of novel immuno-oncology drug, Evotec
AI helps choose a disease target and generate candidate molecules, but lab tests and Phase 1 to 3 trials still decide what works. One candidate credited to generative AI entered Phase 3 in 2026.
AlphaFold is Google DeepMind's AI system for predicting a protein's 3D shape from its sequence. It can do in minutes what took labs months or years, but a prediction is not a medicine.