{"id":12385,"date":"2021-01-05T10:03:47","date_gmt":"2021-01-05T15:03:47","guid":{"rendered":"http:\/\/www.drugpatentwatch.com\/blog\/?p=12385"},"modified":"2026-10-20T10:35:05","modified_gmt":"2026-10-20T14:35:05","slug":"artificial-intelligence-in-drug-discovery-what-is-realistic-what-are-illusions-part-1-ways-to-make-an-impact-and-why-we-are-not-there-yet","status":"publish","type":"post","link":"https:\/\/www.drugpatentwatch.com\/blog\/artificial-intelligence-in-drug-discovery-what-is-realistic-what-are-illusions-part-1-ways-to-make-an-impact-and-why-we-are-not-there-yet\/","title":{"rendered":"AI Drug Discovery: What Works, What Is Hype, and Where the IP Value Sits"},"content":{"rendered":"\n
\"\"<\/figure>\n\n\n\n

On September 14, 2026, the FDA accepted Takeda’s new drug application for zasocitinib under Priority Review. The target action date is in the first quarter of calendar year 2027 [1]. If the FDA approves it, many headlines will call it an AI-designed drug. That label is weak. The molecule came from a ten-year physics-based design program, not from a generative model [3]. The gap between the headline and the record is the main subject of this article.<\/p>\n\n\n\n

I checked the major AI drug discovery claims against primary documents: company press releases, SEC-linked filings, court opinions, USPTO guidance and FDA guidance. I then calculated a few derived metrics from those numbers. Where a figure is my calculation, I label it as analysis. Where I could not verify a claim, I leave it out.<\/p>\n\n\n\n

The Short Answer<\/strong><\/h2>\n\n\n\n

AI works well today for narrow design tasks: predicting protein structure, scoring molecules, optimizing antibodies and proposing targets. It has not yet shown that it raises the odds of Phase 2 and Phase 3 success. I found no FDA approval, as of October 20, 2026, of a drug that its sponsor calls AI-discovered. Two programs are close. Zasocitinib has an NDA under review [1]. Rentosertib started a Phase 3 trial in September 2026 [6].<\/p>\n\n\n\n

The IP value sits in four places: the claims on the molecule, the claims on the new use, the trade secrets in the data and models, and the contracts that move milestone money. It does not sit in a claim that says “we used AI.” US law treats AI as a tool. The inventor must be a human [17][19].<\/p>\n\n\n\n

The Five Findings That Matter<\/strong><\/h2>\n\n\n\n
    \n
  1. Zasocitinib has two Phase 3 trials with 693 and 1,108 patients. In both, the PASI 90 response at week 16 was 61% and 52% for zasocitinib versus 17% and 16% for apremilast [1]. The NDA is under Priority Review.<\/li>\n\n\n\n
  2. Rentosertib moved from a 71-patient, 12-week Phase 2a study to a planned 320-patient, 52-week Phase 3 study. The Phase 3 trial has 47 planned centers, all in China [5][6].<\/li>\n\n\n\n
  3. Upfront cash is a small share of announced AI deal value. In the Isomorphic Labs deals with Lilly and Novartis, upfront payments are about 2.8% of headline value (analysis from [14]). In the 2026 Lilly and Insilico deal, upfront cash is about 4.2% (analysis from [8]).<\/li>\n\n\n\n
  4. Three AI-linked clinical programs ended without efficacy: DSP-1181 in Phase 1, BEN-2293 in Phase 2a, and REC-994 in Phase 2 [10][11][12].<\/li>\n\n\n\n
  5. The USPTO says AI is a tool and only a natural person can invent. The USPTO published that position on November 28, 2025 [19]. The Federal Circuit held the same on the statute in 2022 [17].<\/li>\n<\/ol>\n\n\n\n

    How We Built This Analysis (Methodology)<\/strong><\/h2>\n\n\n\n

    What Data I Used<\/strong><\/h3>\n\n\n\n

    I used company press releases, a peer-reviewed paper summary, regulatory announcements, court decisions summarized by law firms, and USPTO and FDA documents. The period is 2016 to October 20, 2026. I did not use patent counts from commercial databases. DrugPatentWatch lists the approved TYK2 inhibitor in its public class page, but its patent numbers and dates sit behind a subscription [29]. I do not reproduce them.<\/p>\n\n\n\n

    How I Selected Programs<\/strong><\/h3>\n\n\n\n

    I included a program if it met three tests. The sponsor or a primary source links the molecule to AI or computational design. The program has a documented clinical or regulatory event. The event is verifiable from a company, regulator or journal source. I excluded vendor market-size reports and unsourced pipeline counts.<\/p>\n\n\n\n

    How I Calculated Derived Figures<\/strong><\/h4>\n\n\n\n

    I calculated differences and shares from reported numbers. For deal shares, I divided upfront cash by upfront plus stated milestones. Each calculation appears in the text with its inputs. Press releases report headline deal values that include contingent milestones. Those values are not cash paid.<\/p>\n\n\n\n

    Limits of This Analysis<\/strong><\/h4>\n\n\n\n

    Company releases are not peer review. Phase 2a and Phase 3 data from the same sponsor can differ in quality. Two summaries I used are secondary: the Health Tech World summary of the BCG study [16] and the BioPharmaTrend summary of the Nature Medicine paper [5]. I mark those where they appear.<\/p>\n\n\n\n

    What Does “AI-Discovered Drug” Mean?<\/strong><\/h2>\n\n\n\n

    No regulator defines the term. Companies use it for at least four different things. I use the following original classification in this article. It is my framework, not an industry standard.<\/p>\n\n\n\n

    Type<\/th>What the software did<\/th>Example<\/th>Source<\/th><\/tr><\/thead>
    Type 1: Physics-based computational design<\/td>Scored and ranked molecules with free-energy calculations<\/td>Zasocitinib (TAK-279)<\/td>[3]<\/td><\/tr>
    Type 2: AI target discovery plus generative chemistry<\/td>Proposed a new target and generated molecules for it<\/td>Rentosertib (TNIK)<\/td>[6][7]<\/td><\/tr>
    Type 3: AI-engineered biologic<\/td>Optimized an antibody for potency and half-life<\/td>GB-0895 (anti-TSLP)<\/td>[13]<\/td><\/tr>
    Type 4: Knowledge-graph repurposing<\/td>Proposed a new use for an approved drug<\/td>Baricitinib for COVID-19<\/td>[26][27]<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    Why the Label Matters for IP<\/strong><\/h3>\n\n\n\n

    Each type creates a different patent position. A Type 1 or Type 3 program produces a molecule claim. A Type 2 program may also produce a target claim, which is harder to enforce. A Type 4 program produces a method-of-use claim on a drug that another company already owns. I return to this in the IP section.<\/p>\n\n\n\n

    Which AI Drug Discovery Methods Actually Work?<\/strong><\/h2>\n\n\n\n

    Four methods have documented, checkable results. Each result is narrow.<\/p>\n\n\n\n

    Protein Structure Prediction<\/strong><\/h3>\n\n\n\n

    The 2024 Nobel Prize in Chemistry went in part to Demis Hassabis and John Jumper for protein structure prediction, and in part to David Baker for computational protein design [28]. Structure prediction speeds the early work. It does not tell you whether a target is the right one for a disease.<\/p>\n\n\n\n

    Physics-Based Molecular Scoring<\/strong><\/h3>\n\n\n\n

    Nimbus and Schr\u00f6dinger began the TYK2 project in 2016. At the first stage, they listed about 4,000 molecules and used free-energy calculations to score potency, selectivity and solubility. They synthesized 46 and nine met the project criteria [3]. After a pivot to a different protein domain, they screened 2.4 million commercial compounds. They declared the development candidate within two years of that pivot [3]. This is computation doing real work. It is also not generative AI.<\/p>\n\n\n\n

    AI Target Discovery<\/strong><\/h3>\n\n\n\n

    Insilico says its platform linked TNIK to fibrosis. The lead investigator of the Phase 3 trial said in the company’s release that TNIK “had never previously been linked to fibrosis” [6]. That is a company quotation, not an independent finding. The Phase 2a data are the first human test of that idea [5].<\/p>\n\n\n\n

    Antibody Optimization<\/strong><\/h3>\n\n\n\n

    Generate:Biomedicines describes GB-0895 as an antibody engineered with AI. The company says it used machine learning to improve potency and half-life while keeping specificity for TSLP. The goal is a dose every six months [13]. The program has not yet reported Phase 3 results.<\/p>\n\n\n\n

    Case Studies: The Late-Stage Programs<\/strong><\/h2>\n\n\n\n

    Zasocitinib (TAK-279): The Physics-Based Case<\/strong><\/h3>\n\n\n\n

    Takeda paid $4 billion in guaranteed cash for the Nimbus TYK2 program in December 2022 [2]. Schr\u00f6dinger’s own account says Takeda acquired the program by February 2023 and renamed the compound TAK-279 [3]. The compound is a selective inhibitor of the TYK2 JH2 domain [3].<\/p>\n\n\n\n

    Takeda reported its Phase 3 psoriasis data in two trials: LATITUDE PsO 3001 (NCT06088043, 693 participants) and LATITUDE PsO 3002 (NCT06108544, 1,108 participants) [1]. All p-values were below 0.001 against placebo and apremilast at week 16 [1].<\/p>\n\n\n\n

    Week 16 endpoint<\/th>Trial 3001: zasocitinib \/ placebo \/ apremilast<\/th>Trial 3002: zasocitinib \/ placebo \/ apremilast<\/th><\/tr><\/thead>
    sPGA 0\/1<\/td>71% \/ 11% \/ 32%<\/td>69% \/ 13% \/ 30%<\/td><\/tr>
    PASI 75<\/td>76% \/ 12% \/ 37%<\/td>71% \/ 12% \/ 33%<\/td><\/tr>
    PASI 90<\/td>61% \/ 5% \/ 17%<\/td>52% \/ 4% \/ 16%<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    Source: Takeda, September 14, 2026 [1].<\/p>\n\n\n\n

    Analysis: the placebo-adjusted PASI 90 gain is 56 percentage points in trial 3001 and 48 in trial 3002. Against apremilast, the PASI 90 response is 3.6 times higher in trial 3001 and 3.3 times higher in trial 3002. These are my calculations from the reported percentages.<\/p>\n\n\n\n

    The Head-to-Head Trial Against Deucravacitinib<\/strong><\/h4>\n\n\n\n

    On June 11, 2026, Takeda announced LATITUDE Atlas (NCT06973291). It compared zasocitinib 30 mg with deucravacitinib 6 mg in 606 adults. The primary endpoint was PASI 100 at week 16 [4]. Takeda said zasocitinib was statistically superior on the PASI 90 key secondary endpoint. The release text I reviewed gives no percentages and no p-values [4]. I do not report a result I could not see.<\/p>\n\n\n\n

    What Zasocitinib Means for the AI Label<\/strong><\/h4>\n\n\n\n

    The program is ten years old if you count from the 2016 Nimbus and Schr\u00f6dinger start to the 2026 NDA acceptance (analysis from [1][3]). A computer helped choose which molecules to make. People still ran the biology and the trials. I classify it as Type 1.<\/p>\n\n\n\n

    Rentosertib (ISM001-055): The Target Discovery Case<\/strong><\/h3>\n\n\n\n

    Insilico reported Phase 2a results in Nature Medicine on June 3, 2025. The trial, GENESIS-IPF, enrolled 71 patients with idiopathic pulmonary fibrosis across 22 sites in China. Patients received placebo, 30 mg once daily, 30 mg twice daily, or 60 mg once daily for 12 weeks [5]. At 60 mg once daily, the mean FVC change was +98.4 mL. In the placebo group it was -20.3 mL [5]. The trial met its primary safety endpoint [5].<\/p>\n\n\n\n

    Analysis: the gap between the 60 mg arm and placebo is 118.7 mL. With roughly 18 patients per arm, a number I infer from 71 patients and four arms, this is a small sample. The Phase 3 trial will test it.<\/p>\n\n\n\n

    Insilico dosed the first Phase 3 patient on September 10, 2026 [6]. The trial, GENESIS-IPF-3 (NCT07687459), is randomized, double-blind and placebo-controlled. It plans about 320 participants at 47 centers in China. The primary endpoint is the annual rate of FVC decline over 52 weeks [6]. The FDA granted Orphan Drug Designation for IPF in February 2023 [6].<\/p>\n\n\n\n

    What Changed Between Phase 2a and Phase 3<\/strong><\/h4>\n\n\n\n

    Analysis: the sample size grows from 71 to about 320, which is 4.5 times. The treatment period grows from 12 weeks to 52 weeks, which is 4.3 times. The time from Nature Medicine publication to first Phase 3 dosing is about 15 months. I found no statement of a US Phase 3 program in the sources I reviewed. A China-only trial may not meet FDA expectations on its own. That is my reading and not an FDA statement.<\/p>\n\n\n\n

    GB-0895: The Engineered Antibody Case<\/strong><\/h3>\n\n\n\n

    Generate announced two global Phase 3 trials on December 1, 2025: SOLAIRIA-1 and SOLAIRIA-2. They cover about 1,600 patients in more than 40 countries, with a 52-week placebo-controlled design and an annualized exacerbation-rate endpoint. The dose is 300 mg every six months [13]. I found no efficacy data yet.<\/p>\n\n\n\n

    Baricitinib: The Repurposing Case<\/strong><\/h3>\n\n\n\n

    BenevolentAI authors proposed baricitinib for COVID-19 in a February 2020 Lancet letter [27]. The drug was available in the US under Emergency Use Authorization from November 2020. The FDA approved Olumiant for hospitalized adults with COVID-19 on May 11, 2022, based on the ACTT-2 and COV-BARRIER trials [26]. The approval belongs to Lilly and Incyte. Analysis: the AI-assisted idea added a use. It did not add a new composition-of-matter patent for the drug.<\/p>\n\n\n\n

    Where Did AI Drug Discovery Fail?<\/strong><\/h2>\n\n\n\n

    The failures are as informative as the wins. Three are well documented.<\/p>\n\n\n\n

    Program<\/th>Company<\/th>What happened<\/th>Source<\/th><\/tr><\/thead>
    DSP-1181 (OCD)<\/td>Exscientia and Sumitomo Dainippon<\/td>Entered Phase 1 by end of 2020. Development in Japan stopped a year later because the Phase 1 study did not meet expected criteria.<\/td>[12]<\/td><\/tr>
    BEN-2293 (atopic dermatitis)<\/td>BenevolentAI<\/td>Phase 2a, 91 adults, 28 days. Met the safety endpoint. Missed the efficacy endpoints against placebo (April 2023).<\/td>[11]<\/td><\/tr>
    REC-994 (cerebral cavernous malformation)<\/td>Recursion<\/td>Discontinued in May 2025. Phase 2 showed no clear efficacy signal.<\/td>[10]<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    What Recursion’s Numbers Show<\/strong><\/h3>\n\n\n\n

    Recursion closed its Exscientia acquisition in November 2024. At announcement, it expected 10 clinical readouts within 18 months [10]. In May 2025 it dropped REC-994 and REC-2282, weighed out-licensing REC-3964 and paused REC-4539 [10]. Cash fell from $603 million to $509 million in the first quarter of 2025 [10].<\/p>\n\n\n\n

    The second-quarter 2026 report, released August 5, 2026, shows cash of $556.8 million, down from $753.9 million at December 31, 2025. The net loss was $131.0 million, versus $171.9 million a year earlier. Revenue was $7.7 million, versus $19.2 million [9]. Recursion says its runway extends into early 2028 [9].<\/p>\n\n\n\n

    Recursion metric (analysis)<\/th>Inputs<\/th>Result<\/th><\/tr><\/thead>
    Cash change, Q1 2025<\/td>$603M to $509M [10]<\/td>-$94M, -15.6%<\/td><\/tr>
    Cash change, H1 2026<\/td>$753.9M to $556.8M [9]<\/td>-$197.1M, -26.1%<\/td><\/tr>
    Net loss, Q2 2026 vs Q2 2025<\/td>$131.0M vs $171.9M [9]<\/td>-23.8%<\/td><\/tr>
    Revenue, Q2 2026 vs Q2 2025<\/td>$7.7M vs $19.2M [9]<\/td>-59.9%<\/td><\/tr>
    R&D expense, Q2 2026 vs Q2 2025<\/td>$89.6M vs $128.6M [9]<\/td>-30.3%<\/td><\/tr>
    Partner payments received to date<\/td>Genentech $216M plus Sanofi $134M [9]<\/td>$350M combined<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    The pipeline remains active. REC-4881 for familial adenomatous polyposis has Phase 2 data due on November 2, 2026, and an FDA registrational-path update in the second half of 2026 [9]. Early Phase 1 data for REC-617 and REC-3565 are due in the first half of 2027 [9]. The company is smaller and slower than its 2024 plan implied.<\/p>\n\n\n\n

    Do AI-Discovered Drugs Succeed More Often in Clinical Trials?<\/strong><\/h2>\n\n\n\n

    Early data say Phase 1 success is higher. Phase 2 success is about the same as the industry average. The sample is small.<\/p>\n\n\n\n

    \n

    AI-discovered molecules showed an 80-90% Phase 1 success rate, compared with 40-65% for historically discovered molecules. Phase 2 success was about 40%, in line with industry averages. Source: summary of Jayatunga et al., Drug Discovery Today, 2024, by Health Tech World, May 10, 2024 [16].<\/p>\n<\/blockquote>\n\n\n\n

    Why a High Phase 1 Rate Proves Little<\/strong><\/h3>\n\n\n\n

    Phase 1 trials test safety in a small group. A molecule that passes has shown tolerable safety. It has not shown that it works. The Phase 2 figure is the first hard test of efficacy. It sits at about 40%, which the summary calls in line with industry averages. The authors call the work “a first view,” and they note a limited sample for the Phase 2 number [16].<\/p>\n\n\n\n

    What the Failures Add<\/strong><\/h3>\n\n\n\n

    BEN-2293 passed its safety endpoint and failed on efficacy. That is the pattern the Phase 1 versus Phase 2 split predicts [11]. AI can help pick molecules that are safe. The biology still decides whether the disease responds.<\/p>\n\n\n\n

    What Does the Money Say? Deal Economics<\/strong><\/h2>\n\n\n\n

    Big pharma pays for options on AI platforms, not for finished drugs. The upfront share of the headline value shows how much risk the buyer keeps.<\/p>\n\n\n\n

    Deal<\/th>Date<\/th>Upfront<\/th>Stated milestones or total<\/th>Upfront share (analysis)<\/th>Source<\/th><\/tr><\/thead>
    Isomorphic Labs and Eli Lilly<\/td>January 2024<\/td>$45M<\/td>Up to $1.7B milestones<\/td>2.6%<\/td>[14]<\/td><\/tr>
    Isomorphic Labs and Novartis<\/td>January 2024<\/td>$37.5M<\/td>Up to $1.2B milestones, three targets<\/td>3.0%<\/td>[14]<\/td><\/tr>
    Insilico and Eli Lilly<\/td>March 31, 2026<\/td>$115M<\/td>About $2.75B total, plus tiered royalties<\/td>4.2%<\/td>[8]<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    Analysis: the two Isomorphic deals together carry $82.5 million upfront against $2.9 billion in milestones, an upfront share of 2.8%. For the Isomorphic rows, I divided upfront by upfront plus milestones. For the Insilico row, I divided $115 million by the stated $2.75 billion total.<\/p>\n\n\n\n

    What Public Markets Paid<\/strong><\/h3>\n\n\n\n

    Insilico listed on the Hong Kong Stock Exchange under Chapter 8.05 and raised HKD 2.277 billion. Its public tranche was oversubscribed 1,427 times. The share price closed at HKD 29.98 on the first trading day, up 24.66%. Lilly was one of 15 cornerstone investors [7]. The company reports 10 programs with IND clearance and seven in active clinical development [7].<\/p>\n\n\n\n

    What Private Markets Paid<\/strong><\/h3>\n\n\n\n

    Isomorphic Labs raised a $2.1 billion Series B on May 12, 2026, led by Thrive Capital, for a total of about $2.6 billion raised. This is as reported by a secondary source [15]. The company had no program in human trials in that report. It expects first clinical trials by the end of 2026. Its founder had earlier targeted the end of 2025 [15].<\/p>\n\n\n\n

    Can an AI Be an Inventor?<\/strong><\/h2>\n\n\n\n

    No. Not in the US, not in the UK, and the Federal Circuit has said why.<\/p>\n\n\n\n

    Thaler v. Vidal (Fed. Cir. 2022)<\/strong><\/h3>\n\n\n\n

    On August 5, 2022, the Federal Circuit held that “the Patent Act requires that inventors must be natural persons; that is, human beings” [17]. The panel was Chief Judge Moore and Judges Taranto and Stark. Judge Stark wrote the opinion. The court did not decide whether inventions made by humans with AI help are patentable [17].<\/p>\n\n\n\n

    UK Supreme Court: Thaler v Comptroller<\/strong><\/h3>\n\n\n\n

    On December 20, 2023, the UK Supreme Court ruled under section 7 of the Patents Act 1977 that an inventor must be a human. It also left open how much human input is needed where AI helped [18]. The court noted that the result “might well have been different” if Dr. Thaler had claimed to be the inventor and used his system as a tool [18]. I treat that as a quotation of the court’s reasoning as reported by Gowling WLG.<\/p>\n\n\n\n

    USPTO Revised Guidance (November 28, 2025)<\/strong><\/h3>\n\n\n\n

    On November 28, 2025, the USPTO published revised inventorship guidance. It rescinded the February 13, 2024 guidance. It treats AI as a tool, like “laboratory equipment, computer software, research databases” [19]. The test is human conception. The inventor must hold a definite and permanent idea of the complete and operative invention [19]. The Pannu factors do not apply when one human inventor used AI, because AI cannot be a joint inventor [19].<\/p>\n\n\n\n

    What This Means for a Drug Discovery Team<\/strong><\/h4>\n\n\n\n

    Record who chose the target, who picked the molecule, and who understood how to make it. The specification should show that a human had a complete mental picture of the claimed invention at filing [19]. Logs of model runs do not replace that record.<\/p>\n\n\n\n

    Date<\/th>Event<\/th>Forum<\/th>Source<\/th><\/tr><\/thead>
    Aug. 5, 2022<\/td>Inventor must be a natural person<\/td>Federal Circuit, Thaler v. Vidal<\/td>[17]<\/td><\/tr>
    Dec. 20, 2023<\/td>AI cannot be the inventor; human-input question left open<\/td>UK Supreme Court [2023] UKSC 49<\/td>[18]<\/td><\/tr>
    Apr. 18, 2025<\/td>Generic machine learning applied to new data is ineligible<\/td>Federal Circuit, Recentive v. Fox<\/td>[20]<\/td><\/tr>
    Sept. 26, 2025<\/td>ML training claims eligible when they improve the model<\/td>USPTO Appeals Review Panel, Ex parte Desjardins<\/td>[21]<\/td><\/tr>
    Nov. 4, 2025<\/td>Desjardins designated precedential<\/td>USPTO<\/td>[21]<\/td><\/tr>
    Nov. 28, 2025<\/td>Revised AI-assisted inventorship guidance<\/td>USPTO<\/td>[19]<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    Are AI Models and Methods Patentable?<\/strong><\/h2>\n\n\n\n

    Sometimes. The answer depends on whether the claim improves the model or only uses it.<\/p>\n\n\n\n

    Recentive Analytics v. Fox (Fed. Cir. 2025)<\/strong><\/h3>\n\n\n\n

    On April 18, 2025, the Federal Circuit held that claims that apply generic machine learning to new data environments, without improving the models, are ineligible under Section 101. The court said such claims “do no more than claim the application of generic machine learning to new data environments” [20]. It also said machine learning “may lead to patent-eligible improvements in technology” and did not rule on claims that improve the models [20].<\/p>\n\n\n\n

    Ex parte Desjardins (USPTO 2025)<\/strong><\/h3>\n\n\n\n

    On September 26, 2025, the USPTO Appeals Review Panel reversed a Section 101 rejection of continual-learning claims. It found the claims integrated the abstract idea into a practical application by “improving the functioning of machine learning models themselves.” The panel still rejected the claims as obvious [21]. The USPTO designated the decision precedential on November 4, 2025 [21].<\/p>\n\n\n\n

    What This Means for Drug Discovery Platforms<\/strong><\/h4>\n\n\n\n

    A claim like “use a neural network to predict binding for a protein” looks like Recentive. A claim to a new training method that makes the model work better looks like Desjardins. Analysis: the first type is a weak asset. The second type is better but still faces obviousness. Neither claim protects the drug. Only a molecule or use claim does that.<\/p>\n\n\n\n

    Will AI-Generated Molecule Claims Survive Enablement and Written Description?<\/strong><\/h2>\n\n\n\n

    This is the largest legal risk I see for AI-generated genus claims. A model can generate millions of virtual compounds. A patent applicant who claims all of them must still teach the public how to make and use all of them.<\/p>\n\n\n\n

    Amgen v. Sanofi (2023)<\/strong><\/h3>\n\n\n\n

    In May 2023, the Supreme Court unanimously held Amgen’s antibody genus claims not enabled. Amgen had disclosed 26 antibody sequences and claimed a broad class. Justice Gorsuch wrote that the more a party claims, the more it must enable [22]. The Court said Amgen’s two suggested routes gave skilled artisans research tasks, not enablement [22].<\/p>\n\n\n\n

    Duke University v. Sandoz (Fed. Cir. 2025)<\/strong><\/h3>\n\n\n\n

    On November 18, 2025, the Federal Circuit held claim 30 of U.S. Patent No. 9,579,270 invalid for lack of written description. It reversed a $39 million jury verdict [23]. A genus claim needs either a representative number of species or structural features common to all members. The court found neither. It also found too few “blaze marks” to lead a skilled person to the claimed subgenus [23].<\/p>\n\n\n\n

    What This Means for Generative Chemistry<\/strong><\/h4>\n\n\n\n

    Analysis: a generative model creates a large library fast. A patent that claims the whole library on the strength of a few tested examples invites the Amgen and Duke arguments. The safer path is to claim the molecule that was made and tested, plus a narrow set of close analogs with data.<\/p>\n\n\n\n

    Where Does the IP Value Sit?<\/strong><\/h2>\n\n\n\n

    The value sits where an infringement case or an exclusivity period can protect revenue. I rank the layers below. This is my original framework.<\/p>\n\n\n\n

    Layer<\/th>What it protects<\/th>Strength today<\/th>Main risk<\/th><\/tr><\/thead>
    1. Molecule (composition of matter)<\/td>The drug itself<\/td>Strongest<\/td>Genus overreach: enablement and written description [22][23]<\/td><\/tr>
    2. New use or method of treatment<\/td>An indication or dosing for a known drug<\/td>Moderate<\/td>Owner of the base drug controls the market (baricitinib) [26]<\/td><\/tr>
    3. Regulatory exclusivity<\/td>NCE: 5 years. Orphan: 7 years.<\/td>Strong, independent of patents<\/td>Applies only after approval [25]<\/td><\/tr>
    4. Target or biology claims<\/td>Use of a new target<\/td>Weak to moderate<\/td>Broad functional claims face enablement limits [22]<\/td><\/tr>
    5. Model and training-method claims<\/td>The software method<\/td>Weak to moderate<\/td>Section 101 [20][21]<\/td><\/tr>
    6. Data and trade secrets<\/td>Training data, assay results, model weights<\/td>Depends on secrecy<\/td>No patent term; leaks and staff moves<\/td><\/tr>
    7. Deal contracts<\/td>Milestones and royalties<\/td>Contractual<\/td>Contingent on trial success [8][14]<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    Why Regulatory Exclusivity Treats AI Drugs Like Any Other Drug<\/strong><\/h3>\n\n\n\n

    The FDA grants new chemical entity exclusivity for five years and orphan exclusivity for seven years. Exclusivity “may run concurrently with a patent but need not do so” [25]. The page I used was current as of February 2016. These periods do not depend on how the drug was found. An AI-found molecule gets the same clock. Rentosertib already holds an orphan designation for IPF [6].<\/p>\n\n\n\n

    What DrugPatentWatch Shows for the TYK2 Class<\/strong><\/h3>\n\n\n\n

    DrugPatentWatch is a database that tracks drug patents, exclusivities and litigation. Its public TYK2 inhibitor class page lists one approved product: Sotyktu (deucravacitinib), NDA 214958, approved September 9, 2022 [29]. The patent numbers and expiration dates on that page require a subscription [29]. Zasocitinib has no Orange Book entry yet. Orange Book listings follow approval. That is my reading of the process, and I did not find a listing. After any approval, a team can check the listed patents in DrugPatentWatch and compare them with the court record and the Orange Book. A database is a starting point for that check. It is not a substitute for the patent document.<\/p>\n\n\n\n

    Why Data and Trade Secrets Matter More Than Many Expect<\/strong><\/h3>\n\n\n\n

    A patent expires. A well-kept training set can last longer. Companies such as Recursion say their partnerships depend on this kind of asset. Recursion reports $350 million received from Genentech and Sanofi to date [9]. I cannot tell from public records how much of that value comes from data and how much from the molecules.<\/p>\n\n\n\n

    How Does the FDA Treat AI in Drug Discovery?<\/strong><\/h2>\n\n\n\n

    The FDA published a draft guidance on January 6, 2025: “Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products.” It is non-binding [24]. It applies to AI models that affect patient safety, drug quality, or the reliability of nonclinical and clinical results [24]. A firm that uses AI for discovery but relies on traditional processes for safety, quality and stability may not need to change its AI governance, according to Foley’s summary [24].<\/p>\n\n\n\n

    The Credibility Framework<\/strong><\/h3>\n\n\n\n

    The draft describes a sequence: define the question, define the context of use, assess risk, describe the model, describe the data, describe training and describe evaluation. Risk depends on how much the model influences the decision and how serious the consequence is [24]. A trial-analysis model is more exposed than a screening model.<\/p>\n\n\n\n

    What This Means for Sponsors<\/strong><\/h4>\n\n\n\n

    A sponsor that used AI only to choose molecules faces little new FDA burden in the sources I reviewed. A sponsor that uses AI to analyze trial data or to run manufacturing faces more.<\/p>\n\n\n\n

    What Happens Next: Timeline and Scenarios<\/strong><\/h2>\n\n\n\n

    Dated Events to Watch<\/strong><\/h3>\n\n\n\n
    Date<\/th>Event<\/th>Status<\/th>Source<\/th><\/tr><\/thead>
    Nov. 2, 2026<\/td>REC-4881 additional Phase 2 data at CGA-IGC meeting<\/td>Scheduled by company<\/td>[9]<\/td><\/tr>
    End of 2026<\/td>Isomorphic Labs first clinical trials<\/td>Company target, reported; slipped from end of 2025<\/td>[15]<\/td><\/tr>
    Q1 2027<\/td>FDA action date for zasocitinib<\/td>Announced PDUFA target<\/td>[1]<\/td><\/tr>
    First half 2027<\/td>REC-617 and REC-3565 early Phase 1 data<\/td>Company guidance<\/td>[9]<\/td><\/tr>
    52 weeks after enrollment<\/td>GENESIS-IPF-3 FVC endpoint<\/td>Trial under way; no result date stated<\/td>[6]<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    Scenario Analysis<\/strong><\/h3>\n\n\n\n

    These scenarios use identifiable assumptions. They are not forecasts.<\/p>\n\n\n\n

    Scenario A: Zasocitinib Is Approved on Time<\/strong><\/h4>\n\n\n\n

    Assumption: the FDA acts by its Priority Review target in the first quarter of 2027 and approves. Result: a drug from a computational design program reaches the market. The press calls it AI-designed. Takeda’s patents and NCE exclusivity apply as for any drug. The claims would read on the molecule, not on the software.<\/p>\n\n\n\n

    Scenario B: Rentosertib Confirms the Phase 2a Signal<\/strong><\/h4>\n\n\n\n

    Assumption: the 52-week FVC decline is lower than placebo by a clear margin. Result: this would be the first Phase 3 confirmation of an AI-proposed target. I found no source that says this has happened before. The target claim would then matter for competitors.<\/p>\n\n\n\n

    Scenario C: Rentosertib Misses<\/strong><\/h4>\n\n\n\n

    Assumption: the Phase 3 trial shows no benefit over placebo. Result: the Phase 1 versus Phase 2 pattern in the BCG-based data would repeat [16]. The AI target-discovery claim would weaken. Deals with milestone-heavy structures would pay little [8].<\/p>\n\n\n\n

    What This Means for Brand Manufacturers, Generic Entry and Investors<\/strong><\/h2>\n\n\n\n

    What This Means for Brand Manufacturers<\/strong><\/h3>\n\n\n\n

    File on the molecule you made and tested. Name human inventors with clear conception records. Keep model and data assets as trade secrets. Do not rely on a “generated by AI” story to widen claims.<\/p>\n\n\n\n

    What This Means for Generic Entry<\/strong><\/h3>\n\n\n\n

    AI-designed small molecules have the same generic entry path as other small molecules. The ANDA filer looks at the Orange Book listings and the claims. Broad genus claims from generative libraries are targets for invalidity arguments under Amgen and Duke [22][23].<\/p>\n\n\n\n

    What This Means for Investors and Licensors<\/strong><\/h3>\n\n\n\n

    Read the upfront share. A deal with a 3% upfront share transfers little cash and a lot of option value [8][14]. Check which layer of the IP table the asset sits in. A platform deal is not the same as a molecule deal.<\/p>\n\n\n\n

    How Do You Read an AI Drug Claim Without Being Fooled?<\/strong><\/h2>\n\n\n\n

    Use four checks. Each one takes a few minutes with primary sources.<\/p>\n\n\n\n

    Check 1: Which Type of AI Contribution Is It?<\/strong><\/h3>\n\n\n\n

    Place the program in the four-type table above. Zasocitinib is Type 1. Rentosertib is Type 2. A press release that says “AI-discovered” without naming the step the software did gives you no information about the risk.<\/p>\n\n\n\n

    Check 2: What Is the Phase and the Sample?<\/strong><\/h3>\n\n\n\n

    A Phase 2a result in 71 patients over 12 weeks is a signal. A Phase 3 result in several hundred patients over 52 weeks is evidence [5][6]. Companies often describe both in the same words.<\/p>\n\n\n\n

    Check 3: Who Owns the Claim That Matters?<\/strong><\/h3>\n\n\n\n

    For a repurposed drug, the owner of the base drug holds the commercial position [26]. For a new molecule, check the inventors, the assignee and the claims. For a platform company, check which assets it owns outright and which it licenses to partners.<\/p>\n\n\n\n

    Check 4: How Much Cash Moved?<\/strong><\/h3>\n\n\n\n

    A headline of $2.75 billion with $115 million upfront is a 4.2% cash deal (analysis from [8]). The rest depends on trial success, regulatory approval and sales. Read the structure before you read the number.<\/p>\n\n\n\n

    Key Takeaways<\/strong><\/h2>\n\n\n\n
      \n
    • Zasocitinib’s NDA is under Priority Review with a first-quarter 2027 target date. Its design program began in 2016 and used physics-based computation [1][3].<\/li>\n\n\n\n
    • Rentosertib’s Phase 2a showed +98.4 mL FVC at 60 mg versus -20.3 mL for placebo in 71 patients. The Phase 3 trial began dosing on September 10, 2026 [5][6].<\/li>\n\n\n\n
    • Early data show 80-90% Phase 1 success and about 40% Phase 2 success for AI-discovered molecules. The sample is small [16].<\/li>\n\n\n\n
    • DSP-1181, BEN-2293 and REC-994 each ended without efficacy [10][11][12].<\/li>\n\n\n\n
    • Upfront cash is 2.6% to 4.2% of headline value in three AI deals (analysis from [8][14]).<\/li>\n\n\n\n
    • US law treats AI as a tool. Inventors must be natural persons [17][19].<\/li>\n\n\n\n
    • Claims that merely apply generic machine learning are ineligible. Claims that improve the model can qualify [20][21].<\/li>\n\n\n\n
    • Broad genus claims from large generated libraries face enablement and written description limits [22][23].<\/li>\n<\/ul>\n\n\n\n

      FAQ<\/strong><\/h2>\n\n\n\n

      Has the FDA approved an AI-discovered drug?<\/strong><\/h3>\n\n\n\n

      I found no approval as of October 20, 2026 of a drug that its sponsor describes as AI-discovered. Zasocitinib has an NDA under Priority Review with a first-quarter 2027 target date [1]. Whether it counts depends on the definition used.<\/p>\n\n\n\n

      Is zasocitinib an AI-designed drug?<\/strong><\/h3>\n\n\n\n

      It came from Nimbus and Schr\u00f6dinger’s physics-based computational platform, which began in 2016 [3]. It is a computational design success. It is not a generative AI result.<\/p>\n\n\n\n

      What is the status of rentosertib?<\/strong><\/h3>\n\n\n\n

      Phase 3 dosing began on September 10, 2026 in China. The trial plans about 320 participants at 47 centers [6].<\/p>\n\n\n\n

      Can I list an AI system as an inventor on a US patent?<\/strong><\/h3>\n\n\n\n

      No. Only natural persons can be inventors, under Thaler v. Vidal and the November 2025 USPTO guidance [17][19].<\/p>\n\n\n\n

      Does using AI make my invention unpatentable?<\/strong><\/h3>\n\n\n\n

      No. The USPTO applies the same inventorship test whether or not AI assisted. A human must have conceived the complete invention [19].<\/p>\n\n\n\n

      Can I patent a machine learning model for drug discovery?<\/strong><\/h3>\n\n\n\n

      You can if the claim improves the model. Claims that apply generic machine learning to new data are ineligible under Recentive [20]. The USPTO’s Desjardins decision supports claims that improve the model itself [21].<\/p>\n\n\n\n

      Do AI drugs get extra exclusivity?<\/strong><\/h3>\n\n\n\n

      No source I found grants it. Standard periods apply: five years for a new chemical entity and seven years for orphan drugs [25].<\/p>\n\n\n\n

      Do AI-discovered drugs pass clinical trials more often?<\/strong><\/h3>\n\n\n\n

      Phase 1 success was 80-90% in an early analysis. Phase 2 success was about 40%, in line with industry averages [16].<\/p>\n\n\n\n

      How much cash do AI drug deals pay upfront?<\/strong><\/h3>\n\n\n\n

      In the three deals I examined, the upfront cash was 2.6% to 4.2% of headline value (analysis from [8][14]).<\/p>\n\n\n\n

      Does the FDA require sponsors to disclose AI used in discovery?<\/strong><\/h3>\n\n\n\n

      The January 2025 draft guidance is non-binding. It focuses on AI that affects safety, quality or the reliability of study results. A firm using AI only for discovery may not need to change its governance, per one law firm summary [24].<\/p>\n\n\n\n

      References<\/strong><\/h2>\n\n\n\n
        \n
      1. Takeda Pharmaceutical Company. (2026, September 14). FDA accepts NDA for zasocitinib (TAK-279) under Priority Review.<\/em> https:\/\/www.takeda.com\/newsroom\/newsreleases\/2026\/fda-priority-review-zasocitinib-psoriasis\/<\/li>\n\n\n\n
      2. Fidler, B. (2025, December 18). Takeda’s zasocitinib psoriasis results. BioPharma Dive.<\/em> https:\/\/www.biopharmadive.com\/news\/takeda-zasocitinib-tyk2-psoriasis-drug-results-nimbus\/808237\/<\/li>\n\n\n\n
      3. Schr\u00f6dinger. (n.d.). Design of a highly selective, allosteric, picomolar TYK2 inhibitor using novel FEP+ strategies.<\/em> https:\/\/www.schrodinger.com\/design-of-a-highly-selective-allosteric-picomolar-tyk2-inhibitor-using-novel-fep-strategies\/<\/li>\n\n\n\n
      4. Takeda Pharmaceutical Company. (2026, June 11). Zasocitinib outperforms deucravacitinib in head-to-head Phase 3 study.<\/em> https:\/\/takeda.com\/newsroom\/newsreleases\/2026\/zasocitinib-outperforms-deucravacitinib-study\/<\/li>\n\n\n\n
      5. BioPharmaTrend. (2025, June 3). AI-designed TNIK inhibitor shows lung function gains in IPF (summary of Nature Medicine Phase IIa results). https:\/\/www.biopharmatrend.com\/post\/1282-ai-designed-tnik-inhibitor-shows-lung-function-gains-in-ipf<\/li>\n\n\n\n
      6. Insilico Medicine. (2026, September 10). Insilico Medicine doses first patient in GENESIS-IPF-3.<\/em> https:\/\/insilico.com\/news\/isn1009261-insilico-medicine-doses-first-patient-genesis-ipf-3<\/li>\n\n\n\n
      7. Nasdaq. (2025, December 30). Insilico Medicine debuts on HKEX as first AI-driven biotech under Chapter 8.05. https:\/\/www.nasdaq.com\/articles\/insilico-medicine-debuts-hkex-first-ai-driven-biotech-under-chapter-805<\/li>\n\n\n\n
      8. Insilico Medicine. (2026, March 31). Insilico Medicine announces global R&D collaboration with Lilly.<\/em> Pharma Focus Europe. https:\/\/www.pharmafocuseurope.com\/pressreleases\/insilico-medicine-announces-global-rd<\/li>\n\n\n\n
      9. Recursion Pharmaceuticals. (2026, August 5). Recursion reports second quarter 2026 financial results.<\/em> https:\/\/ir.recursion.com\/news-releases\/news-release-details\/recursion-reports-second-quarter-financial-results-genentech<\/li>\n\n\n\n
      10. Fierce Biotech. (2025, May 5). Several months after Exscientia merger, AI biotech outfit Recursion reworks pipeline. https:\/\/www.fiercebiotech.com\/biotech\/several-months-after-exscientia-merge-ai-outfit-recursion-reworks-pipeline<\/li>\n\n\n\n
      11. Taylor, N. P. (2023, April 5). BenevolentAI, cruel R&D: AI-enabled drug flunks midphase eczema trial. Fierce Biotech.<\/em> https:\/\/www.fiercebiotech.com\/biotech\/benevolentai-cruel-rd-ai-enabled-drug-flunks-midphase-eczema-trial-dent-deal-plans<\/li>\n\n\n\n
      12. Pharmaceutical Technology. (n.d.). Has AI peaked in drug discovery and development? https:\/\/www.pharmaceutical-technology.com\/analyst-comment\/ai-peak-drug-discovery-development\/<\/li>\n\n\n\n
      13. Generate:Biomedicines. (2025, December 1). Generate:Biomedicines to initiate global Phase 3 studies of GB-0895.<\/em> Kyodo News PR Wire. https:\/\/kyodonewsprwire.jp\/release\/202512010151<\/li>\n\n\n\n
      14. pharmaphorum. (2024, January 8). Isomorphic signs up Lilly, Novartis for $3bn AI drug hunt. https:\/\/pharmaphorum.com\/news\/isomorphic-signs-lilly-novartis-3bn-ai-drug-hunt<\/li>\n\n\n\n
      15. Noqta. (2026, May). Isomorphic Labs $2 billion AI drug discovery Series B 2026. https:\/\/noqta.tn\/en\/news\/isomorphic-labs-2-billion-ai-drug-discovery-series-b-2026<\/li>\n\n\n\n
      16. Health Tech World. (2024, May 10). AI-discovered drugs achieved higher success rate than those by humans, study. https:\/\/www.htworld.co.uk\/news\/ai\/ai-discovered-drugs-achieved-higher-success-rate-than-those-by-humans-study\/ (summarizing Jayatunga et al., 2024, Drug Discovery Today<\/em>)<\/li>\n\n\n\n
      17. Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022). https:\/\/www.bitlaw.com\/source\/cases\/patent\/Thaler.html<\/li>\n\n\n\n
      18. Gowling WLG. (2024). Patent applications must identify human inventors (discussing Thaler v Comptroller-General of Patents, [2023] UKSC 49). https:\/\/gowlingwlg.com\/en\/insights-resources\/articles\/2024\/patent-applications-must-identify-human-inventors<\/li>\n\n\n\n
      19. Jones Day. (2025, December). USPTO revises guidance on AI-assisted inventorship (discussing Revised Inventorship Guidance for AI-Assisted Inventions, Fed. Reg. Doc. 2025-21457, Nov. 28, 2025). https:\/\/www.jonesday.com\/en\/insights\/2025\/12\/uspto-revises-guidance-on-aiassisted-inventorship<\/li>\n\n\n\n
      20. Winston & Strawn. (2025, April 21). Recentive Analytics v. Fox: precedential case of first impression on machine learning and Section 101. https:\/\/www.winston.com\/en\/insights-news\/recentive-analytics-v-foxin-precedential-case-of-first-impression-federal-circuit-holds-machine-learning-patents-that-do-not-improve-the-technology-ineligible-under-section-101<\/li>\n\n\n\n
      21. IP Update. (2025). Precedential shift: USPTO clarifies patentability of AI training methods (Ex parte Desjardins, Appeal No. 2024-000567). https:\/\/www.ipupdate.com\/?p=7500<\/li>\n\n\n\n
      22. Dinsmore & Shohl. (2023, June 7). The more you claim, the more you must enable: SCOTUS delivers Amgen v. Sanofi opinion. https:\/\/www.dinsmore.com\/publications\/the-more-you-claim-the-more-you-must-enable-scotus-delivers-amgen-v-sanofi-opinion\/<\/li>\n\n\n\n
      23. Haug Partners. (2025). Federal Circuit overturns $39M jury verdict, reinforcing strict written description requirements for genus claims (Duke University v. Sandoz Inc., No. 24-1078, Fed. Cir. Nov. 18, 2025). https:\/\/haugpartners.com\/article\/federal-circuit-overturns-39m-jury-verdict-reinforcing-strict-written-description-requirements-for-genus-claims\/<\/li>\n\n\n\n
      24. Foley & Lardner. (2025, January 15). AI in drug development: FDA releases draft guidance. https:\/\/www.foley.com\/de\/insights\/publications\/2025\/01\/ai-drug-development-fda-releases-draft-guidance\/<\/li>\n\n\n\n
      25. U.S. Food and Drug Administration. (2016). How can I better understand patents and exclusivity? https:\/\/www.fda.gov\/industry\/fda-basics-industry\/how-can-i-better-understand-patents-and-exclusivity<\/li>\n\n\n\n
      26. Eli Lilly and Incyte. (2022, May 11). FDA approves Lilly and Incyte’s Olumiant (baricitinib) for certain hospitalized patients with COVID-19. BioSpace. https:\/\/www.biospace.com\/fda-approves-lilly-and-incyte-s-olumiant-baricitinib-for-the-treatment-of-certain-hospitalized-patients-with-covid-19<\/li>\n\n\n\n
      27. Richardson, P., et al. (2020). Baricitinib as potential treatment for 2019-nCoV acute respiratory disease. The Lancet, 395<\/em>(10223), e30-e31. https:\/\/doi.org\/10.1016\/S0140-6736(20)30304-4<\/li>\n\n\n\n
      28. The Nobel Prize. (2024). The Nobel Prize in Chemistry 2024: Press release.<\/em> https:\/\/www.nobelprize.org\/prizes\/chemistry\/2024\/press-release\/<\/li>\n\n\n\n
      29. DrugPatentWatch. (2026). Drug class: Tyrosine kinase 2 inhibitor.<\/em> https:\/\/www.drugpatentwatch.com\/p\/drug-class\/Tyrosine+Kinase+2+Inhibitor<\/li>\n<\/ol>\n\n\n\n

        <\/p>\n","protected":false},"excerpt":{"rendered":"

        On September 14, 2026, the FDA accepted Takeda’s new drug application for zasocitinib under Priority Review. The target action date […]<\/p>\n","protected":false},"author":1,"featured_media":39971,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_lmt_disableupdate":"","_lmt_disable":"","site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[10],"tags":[],"class_list":["post-12385","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-insights"],"modified_by":"DrugPatentWatch","_links":{"self":[{"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/posts\/12385","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/comments?post=12385"}],"version-history":[{"count":0,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/posts\/12385\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/media\/39971"}],"wp:attachment":[{"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/media?parent=12385"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/categories?post=12385"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/tags?post=12385"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}