{"id":38733,"date":"2026-06-11T10:32:00","date_gmt":"2026-06-11T14:32:00","guid":{"rendered":"https:\/\/www.drugpatentwatch.com\/blog\/?p=38733"},"modified":"2026-05-04T17:05:53","modified_gmt":"2026-05-04T21:05:53","slug":"ai-prior-art-search-how-to-invalidate-key-drug-patents-before-you-pay","status":"publish","type":"post","link":"https:\/\/www.drugpatentwatch.com\/blog\/ai-prior-art-search-how-to-invalidate-key-drug-patents-before-you-pay\/","title":{"rendered":"AI Prior Art Search: How to Invalidate Key Drug Patents Before You Pay"},"content":{"rendered":"\n
\"\"<\/figure>\n\n\n\n

Every pharmaceutical acquisition has a number at its core that almost nobody states plainly: the price you are paying for time. Not the compound, not the clinical data, not the manufacturing infrastructure. Time \u2014 specifically, the years of market exclusivity that the target’s patent portfolio is supposed to deliver. Pfizer paid $43 billion for Seagen in 2023. AstraZeneca paid $39 billion for Alexion in 2021. In both cases, a material share of those premiums reflected a single assumption: that the key patents would hold.<\/p>\n\n\n\n

What happens when they don’t?<\/p>\n\n\n\n

The answer is not a hypothetical. AstraZeneca discovered, after closing on Alexion, that its newly acquired complement inhibitor franchise was entangled in patent litigation with Chugai Pharmaceutical over the half-life extension technology that made Ultomiris commercially distinguishable from Soliris. The resolution cost $775 million \u2014 a payment that did not appear in any pre-acquisition headline valuation, because the patent vulnerability was not identified at the right depth during due diligence [1<\/a>]. The Soliris franchise itself later became the subject of a separate lawsuit alleging that five follow-on patents were fraudulent and used to delay biosimilar entry by four years [2<\/a>].<\/p>\n\n\n\n

These are not freak outcomes. They are the predictable result of a due diligence process that relies on patent counsel doing keyword searches in USPTO databases and producing opinion letters that confirm what the seller already knows. What that process misses \u2014 systematically \u2014 is the prior art that predates the patent claims but was never found during prosecution, never cited by the examiner, and never located by the human searchers who were working under time pressure and billing by the hour.<\/p>\n\n\n\n

AI changes what is findable. That is the thesis of this article, and it is narrower and more specific than the general claims you read about artificial intelligence transforming intellectual property. The claim is not that AI is better at reading claim language than a patent attorney. It is that AI is better at finding documents \u2014 across language barriers, across scientific discipline boundaries, across the full scope of non-patent literature \u2014 that a human searcher would never reach. And in pharmaceutical M&A, one document can be worth billions.<\/p>\n\n\n\n

This article is a practitioner’s guide. It explains the mechanics of AI-powered prior art search, the specific failure modes of traditional due diligence it addresses, the platforms and workflows available as of 2025, and the specific considerations that apply when the target’s key patents cover pharmaceutical compositions, formulations, dosing regimens, and biologic mechanisms. It draws on real deal structures, real PTAB data, and the kind of patent intelligence that platforms like DrugPatentWatch<\/strong> make available to analysts who want to model IP risk as a quantitative input rather than a legal checkbox.<\/p>\n\n\n\n

\n\n\n\n

The Actual Problem with Pharma M&A Patent Due Diligence<\/strong><\/h2>\n\n\n\n

Before examining what AI can do, you need a precise picture of what traditional due diligence fails to do. The failure is not effort. M&A patent due diligence in major pharmaceutical transactions routinely involves multiple firms, dozens of attorneys, and months of work. The failure is structural.<\/p>\n\n\n\n

How Traditional Patent Due Diligence Works<\/strong><\/h3>\n\n\n\n

Standard patent due diligence in a pharmaceutical acquisition runs on two parallel tracks. The first is a freedom-to-operate (FTO) analysis that asks whether the target’s products infringe third-party patents. The second is a validity assessment that asks whether the target’s own patents would survive a challenge. The validity assessment is what concerns us here.<\/p>\n\n\n\n

A conventional validity assessment does the following: it takes the claims of each key patent, identifies the claim elements, constructs keyword searches using those elements, runs those searches against USPTO’s patent database and sometimes against PubMed and Google Scholar, reviews the results, and produces an opinion. That opinion typically concludes that the patents are “likely valid” or “valid absent a comprehensive search,” with the latter caveat doing most of the work.<\/p>\n\n\n\n

The problem is the keyword search. Patent claims are written in a specialized register that evolved over decades to create maximum claim scope while surviving prosecution. The compounds, mechanisms, and formulation methods described in those claims were, when the underlying science was being developed, described in the scientific literature using entirely different terminology. A 2024 patent claiming a modified antibody’s pH-dependent binding properties may be anticipated by a 2009 Japanese conference proceeding that describes the identical mechanism using the vocabulary of structural immunology rather than patent claim language. A keyword search will not connect those two documents. A semantic search engine trained on biomedical literature might.<\/p>\n\n\n\n

The Non-Patent Literature Gap<\/strong><\/h3>\n\n\n\n

The deeper structural gap is non-patent literature (NPL). Under 35 U.S.C. \u00a7 102, prior art includes any printed publication, regardless of format. Conference proceedings, academic theses, clinical trial protocols, regulatory submissions, and technical data sheets are all potentially invalidating if they were publicly available before the effective filing date of the challenged claims. They are also extraordinarily difficult to search systematically.<\/p>\n\n\n\n

A competent prior art search for pharmaceutical patents should cover, at minimum: the full corpus of PubMed-indexed literature, chemistry preprints from ChemRxiv, conference proceedings from major medicinal chemistry and biology conferences, patent applications from at least the USPTO, EPO, JPO, CNIPA, and WIPO, and relevant technical documentation from regulatory agencies. Running that search comprehensively using keyword methods would require months of dedicated effort per patent. Under M&A deal timelines \u2014 which routinely compress due diligence to four to eight weeks \u2014 it does not happen.<\/p>\n\n\n\n

The PTAB’s own data makes the NPL gap concrete. In pharmaceutical IPR proceedings, non-patent literature constitutes the primary or co-primary invalidating reference in a significant share of final written decisions that cancel claims [3<\/a>]. This means that the prior art most likely to kill the patents you are paying for is also the prior art least likely to have been found by the humans who assessed those patents before you wrote the check.<\/p>\n\n\n\n

The Prosecution History Problem<\/strong><\/h3>\n\n\n\n

Prosecution history creates a second layer of risk that AI can help quantify. When an applicant makes claim amendments or arguments to distinguish prior art cited by the examiner, those actions create prosecution history estoppel. Claims that were amended to avoid a prior art reference cannot later be interpreted broadly enough to cover what that reference discloses. This is the doctrine of equivalents in reverse.<\/p>\n\n\n\n

In M&A due diligence, the practical consequence is this: a patent whose claims appear broad on paper may have been narrowed to a sliver during prosecution. The acquirer is buying the sliver. An AI system that processes the complete prosecution history, identifies the claim amendments, maps them against the cited references, and extracts the remaining scope of the claims produces a quantitatively more accurate picture of what the acquirer is actually getting than a human reading the face of the issued patent.<\/p>\n\n\n\n

\n\n\n\n

What AI Actually Does Differently<\/strong><\/h2>\n\n\n\n

The technology underlying AI-powered prior art search is sufficiently varied that “AI search” is nearly meaningless as a category description. What matters for pharmaceutical patent work is the specific capabilities that address the gaps described above.<\/p>\n\n\n\n

Semantic Search and Vector Embeddings<\/strong><\/h3>\n\n\n\n

The core capability that distinguishes AI-powered search from keyword search is semantic understanding. In practice, this means encoding documents \u2014 both the target patent’s claims and the potential prior art corpus \u2014 as dense vector representations (embeddings) in a high-dimensional semantic space. Documents that describe the same technical concept using different terminology will have embeddings that are close together in this space, even if they share no keywords.<\/p>\n\n\n\n

As DrugPatentWatch’s technical documentation notes, “a drug patent that describes a compound’s mechanism using the language of 2010 may be prior art to a 2024 patent that describes an identical mechanism using updated scientific terminology” \u2014 and this is precisely what semantic embedding models are designed to surface [4<\/a>]. The practical implication for pharmaceutical M&A is significant. When a biologic patent from 2019 claims a mechanism that was described, using the vocabulary of structural biology rather than IP law, in a 2007 paper from a laboratory in Osaka, keyword search will miss the connection. Cosine similarity between the embeddings of the two documents will catch it.<\/p>\n\n\n\n

The embedding approach works across languages. Multilingual transformer models can encode a Chinese patent application from CNIPA and a U.S. patent claim into the same semantic space, enabling cross-lingual prior art retrieval. For pharmaceutical patents, where meaningful prior art originates in Japanese, German, French, and Chinese technical literature at roughly the same rate as in English, this is not a marginal improvement \u2014 it is a qualitative change in what is searchable.<\/p>\n\n\n\n

Knowledge Graph Approaches<\/strong><\/h3>\n\n\n\n

Some platforms, including IPRally, represent patents as knowledge graphs rather than text documents. In this approach, the technical features of an invention and the relationships between those features are extracted and stored as graph nodes and edges. Prior art search then becomes a graph similarity problem: find patents whose knowledge graphs share structural similarity with the target patent’s graph.<\/p>\n\n\n\n

The advantage for pharmaceutical patents is specificity. A graph-based representation can encode the relationship between a compound’s structural features and its functional claims in a way that a text embedding cannot. This matters when the claim at issue is a composition claim covering a class of compounds defined by a Markush structure \u2014 a common format for pharmaceutical patents that defines claim scope through variable functional groups rather than named compounds. A graph model can represent Markush coverage in a way that lets an analyst identify prior art that anticipates specific members of the claimed class even when no single prior art document discloses the entire class [5<\/a>].<\/p>\n\n\n\n

Large Language Models for Claim Analysis<\/strong><\/h3>\n\n\n\n

Large language models (LLMs) add a qualitative analysis capability that sits on top of the retrieval function. Once a set of candidate prior art references has been identified through semantic search or graph similarity, an LLM can be prompted to perform the legal analysis: does this reference anticipate claim element X under \u00a7 102? Does the combination of references A and B render claim Y obvious under \u00a7 103? Generate a claim chart mapping this reference to each element of the independent claim.<\/p>\n\n\n\n

Platforms like DeepIP automate this claim charting step, producing structured outputs that map retrieved prior art references to specific claim limitations [6<\/a>]. In a due diligence context, this automation does two things. First, it reduces the time required to assess a large patent portfolio from weeks to days. Second, it produces structured, auditable evidence that can be passed directly to outside patent counsel for final legal review \u2014 or, post-closing, used as the basis for an IPR petition.<\/p>\n\n\n\n

The LLM analysis also adds obviousness combination capability. Section 103 obviousness requires demonstrating not just that each element of a claim was known in the prior art, but that a skilled person would have had motivation to combine the relevant references. LLMs can be prompted to generate motivation-to-combine arguments by analyzing the technical logic connecting the references \u2014 arguments that a human attorney would then review and refine before filing. DeepIP specifically markets its automated \u00a7 103 combination generation as a key capability for invalidity searches [6<\/a>].<\/p>\n\n\n\n

PTAB Outcome Prediction<\/strong><\/h3>\n\n\n\n

The PTAB processed 1,737 IPR petitions in fiscal year 2024 [7<\/a>]. Institution rates have ranged between 56% and 67% over the past five years. For pharmaceutical and biotechnology patents specifically, institution rates are substantially higher \u2014 approaching 100% in some technology sub-categories [8<\/a>].<\/p>\n\n\n\n

Machine learning models trained on the corpus of PTAB decisions can produce probabilistic estimates of both institution probability and final written decision outcomes for a given patent. DrugPatentWatch’s PTAB modeling uses input features including technology classification, claim breadth metrics derived from NLP analysis, prosecution history characteristics, the identity and track record of potential petitioners, and the composition of likely judge panels [4<\/a>]. For M&A due diligence, a PTAB outcome model applied to each key patent in the target’s portfolio produces a probabilistic distribution over the effective exclusivity period \u2014 exactly the input needed for a DCF valuation that accounts for IP risk.<\/p>\n\n\n\n

\n

“In the first half of 2024, 71% of PTAB trials resulted in all challenged claims being cancelled. In 2023 the rate was about 68%, up from a low of 55% in 2019.” \u2014 TT Consultants, Patent Invalidation Trends: PTAB Impact & Global Developments<\/em> (2025) [9<\/a>]<\/p>\n<\/blockquote>\n\n\n\n

Those numbers deserve to be stated plainly in deal memos. If a target’s key patent faces an IPR petition post-closing \u2014 a near certainty for any commercially significant pharmaceutical patent after generic or biosimilar manufacturers identify the opportunity \u2014 there is roughly a 70% probability that all challenged claims will be cancelled at the final written decision stage. That probability does not appear in most acquisition fairness opinions. It should.<\/p>\n\n\n\n

\n\n\n\n

The Pharmaceutical Patent Landscape: What You’re Buying and Why It’s Fragile<\/strong><\/h2>\n\n\n\n

To apply AI prior art analysis effectively, you need to understand the taxonomy of pharmaceutical patent claims and the specific invalidity vulnerabilities associated with each type.<\/p>\n\n\n\n

Composition of Matter Patents<\/strong><\/h3>\n\n\n\n

Composition of matter patents \u2014 claiming the compound itself \u2014 are the crown jewel of pharmaceutical IP. A valid composition patent covers all uses of the claimed compound, regardless of indication, and all generics must design around it. These patents are correspondingly difficult to obtain and, in theory, harder to invalidate than secondary patents.<\/p>\n\n\n\n

In practice, composition of matter patents for small molecules are vulnerable to prior art from three sources: combinatorial chemistry libraries published in academic literature before the patent filing date, patents from competing programs that cover structurally similar compounds under Markush claims, and, increasingly, AI-generated compound datasets from early drug discovery programs that were published or otherwise disclosed before filing.<\/p>\n\n\n\n

For biologics, composition of matter patents covering specific antibody sequences are vulnerable to prior art from antibody sequence databases, which may contain records of structurally similar antibodies generated in earlier research programs. The BLAST algorithm and its successors enable rapid comparison of amino acid sequences, and AI platforms that combine sequence comparison with semantic analysis of the functional claims can identify prior art that a keyword search would not surface.<\/p>\n\n\n\n

Formulation Patents<\/strong><\/h3>\n\n\n\n

Formulation patents \u2014 covering specific delivery vehicles, excipient combinations, or administration methods \u2014 are one of the most common forms of pharmaceutical patent evergreening. The underlying active compound is off-patent; the patented claims cover a specific crystalline form, a particular pH range, a stabilizing excipient, or a nanoparticle delivery system.<\/p>\n\n\n\n

These patents are disproportionately vulnerable to prior art from technical literature that the pharma industry tends to overlook: formulation science journals, excipient manufacturer technical bulletins, conference proceedings from pharmaceutical sciences meetings, and graduate theses from formulation science programs. A practitioner’s manual on lyophilization published in 1998 may anticipate a 2015 formulation patent if it discloses the critical stabilization parameters claimed. AI search across non-patent literature is particularly valuable here because the invalidating prior art is often outside the patent databases that traditional searches prioritize.<\/p>\n\n\n\n

Method of Treatment Patents<\/strong><\/h3>\n\n\n\n

Method of treatment patents cover the use of a known compound in a specific therapeutic application or in a specific patient population. They are the standard tool for extending exclusivity on compounds whose composition of matter protection has expired or is expiring.<\/p>\n\n\n\n

The prior art vulnerability here is clinical literature. If the claimed method of treatment was practiced, even anecdotally or in early-stage studies, before the patent filing date, that practice may constitute prior art. AI search tools that process the full PubMed corpus and can cross-reference clinical trial registries, conference abstracts, and case reports against method of treatment claims provide coverage that keyword searches cannot. The invalidating reference may be a 1999 case series published in a specialty journal with a circulation of 200 physicians \u2014 exactly the kind of document that never appears in a keyword patent search but is exactly the kind of document an embedding-based semantic search can surface.<\/p>\n\n\n\n

Dosing Regimen Patents<\/strong><\/h3>\n\n\n\n

Dosing regimen patents are the most straightforward application of prior art analysis because the claimed element \u2014 the specific dose and dosing interval \u2014 is frequently disclosed, sometimes explicitly and sometimes inferably, in earlier clinical literature. AbbVie’s experience with Humira’s patent thicket is instructive: the PTAB invalidated an AbbVie patent covering a specific dosing regimen for Humira, finding it obvious in light of prior art [10<\/a>]. The prior art was clinical literature that existed at the time of filing; it was found and organized into an IPR petition by a legal team with financial motivation to find it. In M&A due diligence, AI can do the same work without the same time pressure.<\/p>\n\n\n\n

Secondary and Process Patents<\/strong><\/h3>\n\n\n\n

Process patents covering manufacturing methods are a common component of pharmaceutical patent portfolios acquired in M&A transactions and are frequently overvalued because they appear to add exclusivity when they may not. A process patent confers market exclusivity only if every commercially viable manufacturing process for the product is covered by the claims. If the prior art discloses an alternative process that produces the same product and is workable at commercial scale, the exclusivity value of the process patent is zero for the acquirer, regardless of whether the patent is valid.<\/p>\n\n\n\n

AI prior art analysis for process patents needs to cover chemical engineering literature, process chemistry conference proceedings, and patent filings from equipment manufacturers and contract development and manufacturing organizations (CDMOs). This is territory that standard pharmaceutical IP due diligence rarely covers comprehensively.<\/p>\n\n\n\n

\n\n\n\n

The PTAB as a Post-Closing Risk Actuator<\/strong><\/h2>\n\n\n\n

Understanding AI prior art search in the M&A context requires understanding the mechanism by which the prior art becomes financially consequential after the deal closes. That mechanism is the IPR proceeding at the PTAB.<\/p>\n\n\n\n

How an IPR Works and Why It Matters for Acquirers<\/strong><\/h3>\n\n\n\n

An inter partes review (IPR) is an administrative trial at the USPTO’s Patent Trial and Appeal Board in which a petitioner \u2014 typically a generic manufacturer, biosimilar developer, or licensing dispute counterparty \u2014 challenges the validity of issued patent claims on the basis of prior art patents or printed publications. The standard of proof is preponderance of the evidence, substantially lower than the clear and convincing evidence standard required in federal district court. PTAB judges are patent attorneys with technical specialization, not generalist district court judges.<\/p>\n\n\n\n

The consequence of this structure is that the PTAB is a systematically more challenger-favorable forum than district court. As DrugPatentWatch’s analysis of PTAB data demonstrates, institution rates for Bio\/Pharma patents remain near 100% even as the overall PTAB institution rate has fallen below 45% [8<\/a>]. The expected invalidation rate in final written decisions runs at 70-80% for pharmaceutical patents. A $5 billion drug whose key patent survives to its 2029 expiry is a very different financial asset from the same patent invalidated at the PTAB in 2026; the delta in NPV can exceed $10 billion [8<\/a>].<\/p>\n\n\n\n

For a pharmaceutical acquirer, the IPR timeline creates a specific risk window. A generic or biosimilar manufacturer has a statutory one-year window from service of a complaint alleging infringement to file an IPR petition. But any third party \u2014 including a hedge fund running a short position, a competitor attempting to clear a market, or a non-practicing entity \u2014 can file an IPR petition at any point during the patent’s life, without being sued first. The practical result is that every commercially significant pharmaceutical patent the acquirer inherits is one IPR petition away from a 70% probability of invalidation.<\/p>\n\n\n\n

The Deal Valuation Implication<\/strong><\/h3>\n\n\n\n

Standard pharmaceutical M&A valuation models the exclusivity period as a deterministic variable: patent expires on date X, generic entry occurs on date Y, revenue cliff follows. A more accurate model treats the exclusivity period as a probability distribution, where the probability of each outcome year is conditioned on the patent’s IPR vulnerability.<\/p>\n\n\n\n

This is not a theoretical refinement. Between 2025 and 2030, an estimated $236 billion in global pharmaceutical revenue is at risk from patent expirations [10<\/a>]. Acquirers who systematically apply probabilistic patent valuation \u2014 using AI prior art analysis to estimate IPR vulnerability before closing \u2014 will consistently outperform those who treat patent expiry as a hard date.<\/p>\n\n\n\n

The arithmetic is straightforward. Take a target drug generating $2 billion annually, protected by a patent portfolio with nominal expiry in 2031. Traditional due diligence assigns a seven-year exclusivity window with high confidence, supporting a DCF value of $10-12 billion at standard pharmaceutical discount rates. An AI prior art analysis that identifies strong invalidating prior art for the two most significant patents \u2014 prior art that an IPR petitioner could use to plausibly argue for cancellation \u2014 reduces the expected exclusivity period. If the analysis yields a 60% probability that the composition of matter patent is invalidated by 2027, the expected exclusivity window shrinks to approximately 3.8 years, and the DCF value drops to $6-7 billion. That difference in valuation changes whether the deal should be done at all, not merely what the acquirer should pay.<\/p>\n\n\n\n

\n\n\n\n

The AI Prior Art Search Workflow for M&A Due Diligence<\/strong><\/h2>\n\n\n\n

A practical AI prior art search workflow for pharmaceutical M&A due diligence runs through six stages. The first three are preparation; the last three are execution and output.<\/p>\n\n\n\n

Stage 1: Patent Portfolio Mapping and Triage<\/strong><\/h3>\n\n\n\n

The first step is generating a complete, structured map of the target’s patent portfolio. This is not simply listing patents \u2014 it is understanding the relationship between them: which patents protect which products, which claims are asserted against which potential competitors, which patents are listed in the FDA’s Orange Book and therefore trigger automatic injunctive relief under Hatch-Waxman, and which are secondary patents covering formulations, dosing regimens, or methods that might be challenged as obvious evergreening.<\/p>\n\n\n\n

DrugPatentWatch provides this portfolio mapping function for FDA-regulated drugs, cross-referencing Orange Book listings against the full USPTO patent database and providing patent expiration timelines, pending patent applications, and active litigation by patent number [11<\/a>]. For a due diligence team assessing a target’s IP position, this represents a significant time reduction compared to manually constructing the portfolio map from raw USPTO data.<\/p>\n\n\n\n

The triage question at this stage is: which patents are load-bearing? In most pharmaceutical portfolios, a small number of patents account for a disproportionate share of exclusivity value. These are typically the composition of matter patents on the key active ingredient, one or two formulation patents covering the commercially dominant dosage form, and any method of treatment patents covering the primary indication. The AI prior art search should go deepest on these patents. Secondary and tertiary patents in the portfolio may not warrant the same depth of analysis if the acquirer’s risk model is already substantially constrained by the primary patent exposure.<\/p>\n\n\n\n

Stage 2: Claim Deconstruction<\/strong><\/h3>\n\n\n\n

For each patent selected for deep analysis, the independent claims must be deconstructed into their constituent limitations. This is the step that translates legal claim language into searchable technical concepts. A composition claim covering “an antibody comprising a heavy chain variable region having at least 95% sequence identity to SEQ ID NO:1 and a light chain variable region having at least 95% sequence identity to SEQ ID NO:2” breaks down into at least four discrete searchable concepts: the antibody structural type, the heavy chain sequence similarity threshold, the light chain sequence similarity threshold, and any additional functional or binding specificity limitations recited in the claim.<\/p>\n\n\n\n

AI systems that perform automatic claim deconstruction \u2014 extracting claim elements and mapping them to technical concepts in biomedical ontologies \u2014 produce a structured input for the semantic search stage that is more systematic and less dependent on the individual attorney’s framing choices than manual deconstruction. The structured output also creates an auditable record of what was searched and why, which matters both for due diligence quality documentation and for any subsequent IPR proceedings.<\/p>\n\n\n\n

Stage 3: Database Selection and Corpus Construction<\/strong><\/h3>\n\n\n\n

Prior art for pharmaceutical patents can exist in any of the following sources: patent databases (USPTO, EPO, WIPO, JPO, CNIPA, KIPO, India Patent Office), academic literature (PubMed, ChemRxiv, bioRxiv, EMBASE, Web of Science, Scopus), conference proceedings (American Chemical Society, American Society for Biochemistry and Molecular Biology, AACR, ASHP, DIA), regulatory documents (FDA briefing documents, EMA assessment reports, clinical study reports submitted in FOIA responses), technical literature from instrument and materials manufacturers, and gray literature including patents from adjacent industries.<\/p>\n\n\n\n

For any specific patent family, not all sources are equally relevant. A composition of matter patent for a small molecule kinase inhibitor warrants deep search in medicinal chemistry literature, combinatorial chemistry patent databases, and kinase biology conference proceedings. A formulation patent warrants coverage of pharmaceutical sciences journals, excipient manufacturer technical documentation, and CDMO patent filings. Selecting and weighting the search corpus is a judgment call that benefits from domain expertise \u2014 which is why AI search tools work best when deployed with pharmaceutical-specialized input, not as black-box search engines applied uniformly across patent types.<\/p>\n\n\n\n

Stage 4: Semantic Retrieval and Ranking<\/strong><\/h3>\n\n\n\n

The semantic retrieval stage is where AI search technology most directly improves on traditional methods. The claim elements extracted in Stage 2 are encoded as query vectors and matched against the corpus documents using cosine similarity or equivalent distance metrics in the embedding space. The top-ranked documents from each query vector are then combined, deduplicated, and re-ranked using a cross-encoder model that scores each candidate document against the full claim text rather than individual elements.<\/p>\n\n\n\n

The platforms most used in pharmaceutical prior art work as of 2025 include: Patlytics, which combines semantic search with automated claim charting; DeepIP, which specializes in invalidity search with automated \u00a7 103 combination generation; IPRally, which uses knowledge graph representations for pharmaceutical and biotech patents; PatentScan, which focuses specifically on invalidity searches and opposition proceedings; and Darts-ip, which maintains a curated database of patent litigation outcomes globally [5<\/a>, 6<\/a>].<\/p>\n\n\n\n

For pharmaceutical work specifically, the Lens.org database deserves mention. It provides open access to the full text of patents from over 95 jurisdictions and links patent citations to academic literature through CrossRef DOI resolution. An AI search system that uses Lens as a backend can retrieve the non-patent literature cited in prosecution histories, then search semantically outward from those citations to find related literature that was not cited \u2014 literature that may constitute stronger prior art than what the examiner found.<\/p>\n\n\n\n

Stage 5: Evidence Assembly and Claim Charting<\/strong><\/h3>\n\n\n\n

Retrieval produces candidates; evidence assembly determines which candidates are legally significant. The distinction matters because a document that superficially resembles a patent claim may not actually disclose each limitation of that claim with the specificity required for anticipation under \u00a7 102. A document that discloses most but not all limitations of a claim is potentially relevant to an obviousness combination under \u00a7 103, but only if there is a plausible motivation to combine it with other references.<\/p>\n\n\n\n

AI claim charting tools automate the initial evidence assembly step by generating element-by-element claim charts that map retrieved documents to claim limitations, annotating the specific passages that support each mapping, and flagging limitations for which no reference was found. This automated charting serves as the input for attorney review rather than the final legal product, but it reduces the attorney review time from weeks to days and ensures that every limitation is addressed rather than selectively analyzed.<\/p>\n\n\n\n

The output format matters for M&A due diligence specifically. Claim charts that can be exported as structured data \u2014 rather than just as PDF documents \u2014 allow the valuation team to build quantitative models directly from the evidence assembly output. A chart showing that eight of ten independent claim limitations have strong prior art coverage, with the remaining two limitations having moderate coverage, supports a probabilistic IPR success estimate that can feed directly into the patent NPV model.<\/p>\n\n\n\n

Stage 6: PTAB Risk Modeling and Valuation Integration<\/strong><\/h3>\n\n\n\n

The final stage translates the prior art evidence into a financial input. This requires combining two probability estimates: the probability that the identified prior art would support a successful IPR institution, and the probability that an IPR that is instituted results in claim cancellation at the final written decision stage.<\/p>\n\n\n\n

For institution probability, the key inputs are the strength of the best prior art reference relative to each independent claim and the technology art unit of the challenged patent. PTAB institution rates are significantly higher in pharmaceutical and biotechnology technology art units than in the overall petition population [7<\/a>]. Machine learning models trained on the historical PTAB record, using the claim breadth metrics and prior art strength indicators as features, produce institution probability estimates that are more accurate than any attorney’s judgment-based estimate.<\/p>\n\n\n\n

For final written decision outcomes, the same models incorporate the full procedural history: has the patent been previously litigated? Are there claim amendments in the prosecution history that constrain claim scope? Has the patent already survived a reexamination? Each of these factors adjusts the probability distribution over outcomes.<\/p>\n\n\n\n

The combined output \u2014 a probability distribution over effective exclusivity periods for each key patent \u2014 integrates directly into a DCF valuation model. The acquirer can then compare the AI-enhanced patent valuation against the deal price and assess whether the premium paid reflects an accurate or inflated exclusivity assumption.<\/p>\n\n\n\n

\n\n\n\n

Case Studies: Where AI Prior Art Would Have Changed the Analysis<\/strong><\/h2>\n\n\n\n

AstraZeneca \/ Alexion: The Half-Life Extension Technology<\/strong><\/h3>\n\n\n\n

When AstraZeneca acquired Alexion in 2021 for $39 billion, the core IP value in the portfolio resided in the complement inhibitor franchise: Soliris (eculizumab) for rare complement-mediated diseases, and Ultomiris (ravulizumab) as the intended successor product with extended dosing intervals [12<\/a>]. The extended dosing interval for Ultomiris \u2014 administered every eight weeks versus Soliris’s every two weeks \u2014 was commercially critical because it represented the primary differentiation argument for patient transition from Soliris to Ultomiris.<\/p>\n\n\n\n

That extended dosing interval depended on half-life extension technology. Chugai Pharmaceutical held patents on antibody engineering methods for extending half-life through pH-dependent binding modifications. Chugai had filed suit in 2018, before the AstraZeneca acquisition, alleging that Ultomiris infringed those patents. The litigation was disclosed; the risk was known. What was apparently not fully resolved in due diligence was the strength of Alexion’s non-infringement and invalidity defenses against those claims.<\/p>\n\n\n\n

The post-closing resolution \u2014 a $775 million payment to Chugai in 2022, representing a single-payment settlement with no ongoing royalties [1<\/a>] \u2014 indicates that Alexion’s position was weaker than the deal price reflected. An AI prior art analysis of the Chugai patents, conducted during due diligence, would have searched the structural biology and antibody engineering literature for prior disclosures of pH-dependent binding modifications affecting antibody half-life. If that analysis had identified strong invalidating art, it would have given AstraZeneca either a negotiating lever (the patents could be challenged) or a more accurate risk estimate (the settlement cost should be modeled into the acquisition price).<\/p>\n\n\n\n

Separately, Alexion’s own Soliris patent portfolio became the subject of litigation in 2025, with insurers alleging that five follow-on patents were fraudulently obtained and used to delay biosimilar entry by four years beyond the original compound patent’s expiry in 2021 [2<\/a>]. Soliris generated $2.5 billion in sales in 2024 before biosimilar competition materialized [2<\/a>]. If the litigation succeeds, the damages exposure flows to AstraZeneca as the post-acquisition patent holder. An AI prior art analysis of the five follow-on patents before the acquisition closed would have assessed their vulnerability \u2014 and might have identified the same prior art that plaintiffs are now using to argue they should never have been granted.<\/p>\n\n\n\n

Pfizer \/ Seagen: ADC Linker Chemistry<\/strong><\/h3>\n\n\n\n

Pfizer acquired Seagen in 2023 for $43 billion. A significant portion of that premium reflected Seagen’s antibody-drug conjugate (ADC) intellectual property, particularly the patents covering linker chemistry and conjugation methods that made its ADC platform commercially viable [12<\/a>]. ADC patents are a particularly interesting case study for AI prior art analysis because the technology draws on both antibody chemistry and cytotoxic payload chemistry, and the prior art landscape spans immunology, organic chemistry, and targeted therapy literature from the late 1980s through the 2000s \u2014 an era of intense research productivity that predates the current ADC clinical applications.<\/p>\n\n\n\n

The ADC field has a well-documented patent dispute history. The fundamental linker chemistry used in most current ADCs has been the subject of multiple IPR petitions and district court validity challenges. For an acquirer paying $43 billion for an ADC platform, the central IP question is not whether the composition of matter patents on Seagen’s specific clinical compounds are valid \u2014 those are more recently filed and harder to anticipate \u2014 but whether the platform patents on the underlying linker and conjugation technology are valid. Platform patent invalidation would not remove the compound patents, but it would eliminate Seagen’s licensing leverage over competitors developing their own ADC programs, significantly reducing the acquired IP estate’s strategic value.<\/p>\n\n\n\n

An AI prior art search focused on the Seagen platform patents would have processed the full corpus of ADC chemistry literature from the 1990s and early 2000s, including technical publications from Immunomedics and Medarex (both early ADC developers with extensive publication records) and conference proceedings from the American Chemical Society and AACR that predate the filing dates of the key platform patents. Whether such a search would have identified strong invalidating art is unknowable without conducting it \u2014 but the question should have been answered before $43 billion changed hands.<\/p>\n\n\n\n

The Humira Patent Thicket: A Retrospective Lesson<\/strong><\/h3>\n\n\n\n

AbbVie’s construction of a patent thicket around Humira (adalimumab) is now the most analyzed case study in pharmaceutical patent strategy. AbbVie filed over 130 patents covering the original molecule, formulations, dosing regimens, methods of treatment, manufacturing processes, and devices \u2014 a portfolio so dense that biosimilar competitors agreed to delayed entry settlements rather than face patent litigation on each of those layers.<\/p>\n\n\n\n

The PTAB chipped away at this thicket systematically. In particular, it invalidated AbbVie patent claims covering specific dosing regimens for Humira, finding them obvious in light of prior clinical literature [10<\/a>]. The clinical literature that constituted the prior art was not obscure \u2014 it was published in major rheumatology journals and available in PubMed. The reason it was not cited during prosecution was not that it was unfindable; it was that the examiner’s search, conducted using keyword methods at the time of prosecution, did not surface it.<\/p>\n\n\n\n

Any acquirer of AbbVie assets in an era before biosimilar competition should have run a comprehensive prior art analysis of the dosing regimen and method of treatment patents \u2014 not because the acquisition necessarily would have been priced differently, but because the IP risk model would have been more accurate. The generic entry timeline for Humira in the U.S. was ultimately shaped more by settlement agreements than by successful IPR invalidation, but that outcome was not predictable in advance from the face of the patent portfolio.<\/p>\n\n\n\n

\n\n\n\n

DrugPatentWatch as a Due Diligence Infrastructure Layer<\/strong><\/h2>\n\n\n\n

Specialized pharmaceutical patent intelligence platforms occupy a specific role in the AI-augmented due diligence stack: they provide the structured pharmaceutical-specific data layer that general-purpose AI patent search tools lack.<\/p>\n\n\n\n

DrugPatentWatch maintains structured data on Orange Book patent listings, patent expiration timelines, active patent challenges including PTAB petitions and Paragraph IV certifications, biosimilar application filings and their associated patent disputes, and the litigation history associated with each patent in the Orange Book [11<\/a>]. For an M&A due diligence team, this data layer answers the question: which of the target’s patents are currently being challenged, by whom, and at what stage?<\/p>\n\n\n\n

That information changes the prior art search priority queue. If a target’s key composition of matter patent already has an IPR petition pending, the petitioner’s claim charts \u2014 filed as part of the IPR record and publicly available \u2014 constitute a map of the prior art that a well-resourced challenger has already identified. The due diligence team does not need to duplicate that search; it needs to assess the strength of the prior art that has already been found and estimate the likelihood of institution and final written decision outcomes. DrugPatentWatch’s PTAB tracking data provides the context needed to make that assessment efficiently.<\/p>\n\n\n\n

For patents that are not yet subject to active challenges, DrugPatentWatch’s exclusivity timeline data enables the due diligence team to model the window of vulnerability: when the product becomes commercially significant enough to attract a biosimilar or generic application, which triggers the patent dance under Hatch-Waxman or the BPCIA, which creates the IPR filing window. For a target drug that is pre-launch, the IP risk period begins 30 to 36 months post-launch when the first Paragraph IV certification or biosimilar application is likely to arrive. Planning the AI prior art analysis around that timeline \u2014 identifying the vulnerabilities before the challengers do \u2014 gives the acquirer options.<\/p>\n\n\n\n

\n\n\n\n

Non-Patent Literature: The Most Undervalued Prior Art Category<\/strong><\/h2>\n\n\n\n

Patent search professionals describe non-patent literature as the category of prior art most likely to kill a pharmaceutical patent and least likely to have been found by traditional searches. This combination makes it the highest-return target for AI search capability.<\/p>\n\n\n\n

Academic Literature and the PubMed Gap<\/strong><\/h3>\n\n\n\n

PubMed indexes approximately 37 million records across biomedical and life sciences literature. A keyword search of PubMed using the terms appearing in a patent claim will systematically miss documents that use synonymous terminology, documents in languages other than English, and documents from adjacent scientific disciplines that do not use the clinical or pharmaceutical vocabulary of the claim language.<\/p>\n\n\n\n

An embedding-based semantic search of PubMed can surface all three categories. The technique works by encoding the patent claim text as a query vector, searching against pre-computed embeddings of the PubMed abstract corpus, and ranking results by semantic similarity. A 2023 evaluation of this approach for pharmaceutical patent prior art found that semantic search surfaced between 40% and 60% more relevant prior art references than keyword search on the same queries \u2014 with the advantage concentrated in non-English and cross-disciplinary literature [13<\/a>].<\/p>\n\n\n\n

The implications for formulation patents are particularly pronounced. Formulation science draws on physical chemistry, polymer science, materials science, and food science literature in addition to pharmaceutical sciences proper. A patent claiming a specific emulsion formulation for an anti-inflammatory drug may be anticipated by a 1994 paper in a polymer science journal that discloses identical emulsion parameters for a different active. Keyword search of PubMed using pharmaceutical terminology will not find that paper. A semantic search that encodes the emulsion chemistry parameters as query vectors and searches across the full PubMed and Web of Science corpus will.<\/p>\n\n\n\n

Conference Proceedings: The Prior Art Archive Nobody Searches<\/strong><\/h3>\n\n\n\n

Conference proceedings are published, publicly accessible, and constitute prior art under \u00a7 102 if distributed before the effective filing date of the challenged claims. They are also systematically excluded from most patent searches because they are not indexed in the major patent databases and are inconsistently indexed in academic literature databases.<\/p>\n\n\n\n

For pharmaceutical patents, three conference archives are particularly important: the American Chemical Society (ACS) national meeting proceedings, which publish abstracts of medicinal chemistry presentations that often predate full journal papers by 12 to 24 months; the AACR annual meeting proceedings for oncology drug patents; and the American Society for Biochemistry and Molecular Biology (ASBMB) proceedings for biologic drug patents.<\/p>\n\n\n\n

Historically, these archives were difficult to search systematically because they existed as physical volumes or PDFs without structured metadata. AI-powered document processing \u2014 which can extract text from PDFs, normalize heterogeneous document formats, and generate embeddings for semantic search \u2014 makes these archives searchable in a way they have not been before. A due diligence team that adds ACS and AACR proceedings to its AI prior art search corpus is accessing prior art territory that no traditional search workflow would cover.<\/p>\n\n\n\n

Regulatory Documents and FOIA-Released Clinical Study Reports<\/strong><\/h3>\n\n\n\n

Regulatory submissions constitute an underexplored category of prior art for method of treatment and dosing regimen patents. FDA briefing documents for advisory committee meetings are publicly available and may disclose clinical data that predates a patent filing and anticipates claimed dosing parameters. EMA assessment reports (the European equivalent of FDA drug reviews) are public documents that provide detailed summaries of clinical trial data, including safety and efficacy data that may disclose the dose-response relationships claimed in secondary patents.<\/p>\n\n\n\n

Clinical study reports released under FOIA or under EMA’s access to clinical data policies contain detailed trial data that can anticipate dosing regimen claims with high specificity. For an acquirer whose target has a portfolio of dosing regimen patents \u2014 a common feature of mature pharmaceutical products seeking extended exclusivity \u2014 systematic review of available regulatory documents using AI text processing can identify anticipating disclosures that would never appear in a patent database search.<\/p>\n\n\n\n

\n\n\n\n

Building the Internal AI Prior Art Capability: Build, Buy, or Hybrid<\/strong><\/h2>\n\n\n\n

M&A due diligence teams evaluating how to deploy AI prior art analysis face a build-buy-hybrid decision that depends on deal frequency, internal IP expertise, and budget allocation.<\/p>\n\n\n\n

The Case for External Platforms<\/strong><\/h3>\n\n\n\n

For most pharmaceutical acquirers \u2014 including large-cap companies doing three to five acquisitions per year \u2014 the right answer is using external platforms for the retrieval and claim charting functions while keeping the valuation integration and final legal review in-house. External platforms like Patlytics, DeepIP, and PatentScan have already built and tuned the embedding models, assembled the patent and literature corpora, and developed the claim charting output formats. Replicating those capabilities internally costs more than the platform subscription fees for any reasonable deal volume, and the external platforms maintain their models and corpora on an ongoing basis rather than requiring internal engineering resources for upkeep.<\/p>\n\n\n\n

The selection criteria among external platforms for pharmaceutical M&A work are four: the coverage and freshness of the non-patent literature corpus, the quality of cross-lingual search capability (specifically Japanese, German, and Chinese), the availability of automated \u00a7 103 combination generation, and the ability to export structured data outputs (not just PDFs) for integration with valuation models. Not every platform excels on all four criteria; the choice should reflect the specific patent types most common in the acquirer’s target profile.<\/p>\n\n\n\n

The Internal Capability Layer<\/strong><\/h3>\n\n\n\n

What should remain in-house is the domain expertise needed to configure the AI search correctly and interpret the outputs. A platform that returns 500 semantic search results is useful only if an internal team or qualified outside counsel can assess which of those 500 results constitutes legally significant prior art for the specific claims under analysis. That assessment requires understanding of pharmaceutical chemistry, claim construction doctrine, and PTAB jurisprudence on obviousness combinations \u2014 a combination of expertise that cannot be automated and should not be outsourced entirely to the platform vendor.<\/p>\n\n\n\n

The internal capability that has the highest ROI for pharmaceutical acquirers is a Patent Intelligence function that sits at the intersection of the M&A team, the IP legal team, and the scientific\/medical affairs team. This function uses DrugPatentWatch and similar platforms to maintain continuous awareness of the patent risk landscape for potential targets, configures and manages AI prior art searches when deals enter due diligence, and translates patent risk assessments into quantitative inputs for deal valuation models.<\/p>\n\n\n\n

Hybrid Workflows in Practice<\/strong><\/h3>\n\n\n\n

In practice, the most effective AI prior art workflow for pharmaceutical M&A due diligence runs as follows: the internal Patent Intelligence function uses DrugPatentWatch to map the target’s patent portfolio and identify the load-bearing patents within the first week of deal access. External AI platforms are then engaged to conduct semantic prior art searches across the identified patents, with search corpora configured based on internal domain expertise. The platform outputs \u2014 ranked candidate references and automated claim charts \u2014 are reviewed by outside patent counsel, who produce abbreviated invalidity opinions on the top three to five patents within the due diligence timeline. Those opinions feed into a probabilistic patent NPV model maintained by the internal finance team, which adjusts the valuation accordingly.<\/p>\n\n\n\n

This workflow compresses the full cycle from patent identification to financial output to three to four weeks \u2014 feasible within a standard M&A due diligence timeline. It produces quantitatively better patent risk estimates than traditional due diligence approaches without requiring the acquirer to build entirely new internal capabilities.<\/p>\n\n\n\n

\n\n\n\n

Limitations of AI Prior Art Analysis<\/strong><\/h2>\n\n\n\n

The case for AI prior art analysis in pharmaceutical M&A is strong, but overstating it creates its own risks. Three limitations deserve explicit acknowledgment.<\/p>\n\n\n\n

Retrieval Precision and False Positives<\/strong><\/h3>\n\n\n\n

Semantic search produces results ranked by vector similarity, not by legal relevance. A high-cosine-similarity document may discuss the same biological mechanism as the claimed invention without disclosing the specific claim elements required for anticipation. False positives in the retrieval stage consume attorney review time and can create a false sense of comprehensive coverage if the review team focuses on the high-similarity false positives and underweights lower-similarity documents that contain the truly relevant disclosures.<\/p>\n\n\n\n

Mitigating this requires calibrating the retrieval system for pharmaceutical patent claim types \u2014 understanding that the similarity threshold that works for small molecule composition claims is different from the threshold that works for biologic sequence claims \u2014 and maintaining sufficiently broad retrieval cutoffs to avoid false negatives even at the cost of including more false positives in the review set.<\/p>\n\n\n\n

Confidential Prior Art<\/strong><\/h3>\n\n\n\n

AI prior art search, however comprehensive, cannot retrieve prior art that was not publicly disclosed. Confidential research programs, proprietary compound libraries, and unpatented trade secrets may constitute prior art under \u00a7 102(b)(1) exceptions or under derivation proceedings, but they are invisible to search. An acquirer whose target’s key patents were based on research that derived from a third party’s confidential program faces a derivation or inequitable conduct risk that no amount of prior art searching will uncover \u2014 that risk requires human intelligence gathering and representations in the acquisition agreement.<\/p>\n\n\n\n

The Legal Opinion Gap<\/strong><\/h3>\n\n\n\n

AI-generated claim charts and prior art analyses are not legal opinions. They are evidentiary inputs that require attorney review, claim construction analysis, and professional judgment about the strength of invalidity arguments before they can support either a deal valuation or a PTAB petition. Acquirers who treat AI platform outputs as definitive validity assessments without attorney review are creating a different kind of risk: the risk that the AI’s interpretation of the claim scope differs materially from the legally correct interpretation, and that the invalidity argument that appeared strong in the AI claim chart would fail at institution because it rested on an erroneous claim construction.<\/p>\n\n\n\n

The workflow should be AI-first for retrieval and initial evidence organization, attorney-final for claim construction and legal opinion. The AI stages reduce cost and improve coverage; the attorney stage ensures legal accuracy and professional accountability.<\/p>\n\n\n\n

\n\n\n\n

The Post-Closing Use Case: Building the IPR Case<\/strong><\/h2>\n\n\n\n

AI prior art analysis is not only a pre-closing due diligence tool. For acquirers who discover post-closing that they have inherited patents with significant vulnerability, it is also the starting point for a post-closing IP remediation strategy.<\/p>\n\n\n\n

Mapping the Vulnerability Before Competitors Do<\/strong><\/h3>\n\n\n\n

The single most important post-closing IP action for a pharmaceutical acquirer is mapping the vulnerability of the inherited patent portfolio before potential challengers do. Generic and biosimilar manufacturers maintain their own prior art search programs and will identify the same vulnerabilities that AI search would surface. The question is timing: if the acquirer identifies the vulnerable patents before an IPR petition is filed, there are options. If the acquirer identifies them after an institution decision has been issued, the options are limited to settlement or defense.<\/p>\n\n\n\n

Pre-emptive vulnerability mapping allows an acquirer to prepare continuation patent applications that attempt to shore up coverage in the claim areas where prior art exists, design around the weak claims in future product formulations, seek inter partes reexamination to attempt to narrow claims before an IPR challenge forces the issue, and identify the strongest patents in the portfolio for prioritized enforcement \u2014 because a licensing program built on patents with strong prior art support is more defensible than one built on patents the acquirer suspects would not survive challenge.<\/p>\n\n\n\n

Settlement Leverage and the IPR Petition Threat<\/strong><\/h3>\n\n\n\n

For acquirers who inherit patents in competitive therapeutic areas, the identified prior art from an AI search also has offensive uses. In licensing negotiations with third parties, the strength or weakness of the acquirer’s own patents changes the negotiating dynamic. An acquirer who knows its own patent is vulnerable to prior art that has been identified but not yet deployed has a different negotiating position \u2014 and a different walk-away threshold \u2014 than an acquirer who believes the patent is unassailable.<\/p>\n\n\n\n

This self-knowledge also affects the decision of whether to assert the patent aggressively. A pharmaceutical company that sues a competitor for infringement of a patent with known prior art vulnerability invites an invalidity counterclaim that could cancel the patent before the litigation resolves in the plaintiff’s favor. Knowing the vulnerability ahead of time allows the company to price that risk into its litigation strategy rather than discovering it in the defendant’s answer.<\/p>\n\n\n\n

\n\n\n\n

Global Considerations: EPO Oppositions and the UPC<\/strong><\/h2>\n\n\n\n

U.S.-focused analysis tends to center on the PTAB, but pharmaceutical M&A routinely involves patent portfolios with global coverage. The prior art analysis workflow described above applies with modifications to European patent opposition proceedings and, increasingly, to the new Unified Patent Court.<\/p>\n\n\n\n

EPO Opposition Procedures<\/strong><\/h3>\n\n\n\n

The European Patent Office allows any party to file an opposition within nine months of a European patent grant. Roughly 2-3% of EPO-granted patents are opposed, but of those, approximately 25% are fully revoked and another 46% are amended [14<\/a>]. For pharmaceutical patents, which are among the more frequently opposed at the EPO, the opposition statistics are relevant to cross-border deal valuations: a European patent that is granted but within its opposition window at the time of acquisition is subject to challenge by any party, without requiring the challenger to have standing in the U.S. sense.<\/p>\n\n\n\n

The prior art for EPO opposition proceedings includes the same NPL categories as U.S. IPR proceedings, with the addition that European patent examiners have traditionally been more aggressive about citing Japanese and Korean literature than their USPTO counterparts \u2014 meaning that the AI prior art search for European patents should specifically weight Asian-language literature sources.<\/p>\n\n\n\n

The Unified Patent Court<\/strong><\/h3>\n\n\n\n

The UPC, which opened in June 2023, created a new forum for centralized patent revocation across participating EU countries. A single UPC revocation action can invalidate a European patent \u2014 or a unitary patent \u2014 across all participating EU member states simultaneously [14<\/a>]. By mid-2024, the UPC had issued its first revocation judgment, and several early UPC revocation actions involve pharmaceutical patents [9<\/a>].<\/p>\n\n\n\n

For pharmaceutical M&A involving European market revenues, the UPC adds a new post-closing risk vector: a third party can file a single central revocation action that eliminates patent protection across the EU in one proceeding. The prior art analysis applicable to UPC revocation actions is structurally identical to the analysis applicable to EPO opposition and U.S. IPR proceedings, but the financial stakes are amplified by the multi-country scope. AI prior art analysis for pharmaceutical M&A due diligence should explicitly account for UPC vulnerability as a distinct risk dimension from U.S. PTAB vulnerability.<\/p>\n\n\n\n

\n\n\n\n

Practical Integration: Building AI Prior Art Into the Deal Process<\/strong><\/h2>\n\n\n\n

The preceding sections describe what AI prior art analysis can do. This section addresses how to integrate it into a deal process that is already compressed, resource-constrained, and operationally demanding.<\/p>\n\n\n\n

The Phase Structure<\/strong><\/h3>\n\n\n\n

Pharmaceutical M&A due diligence runs through identifiable phases: initial screening (pre-LOI, 2-4 weeks), confirmatory due diligence (post-LOI, 4-8 weeks), and final negotiation (post-commitment, 2-4 weeks). AI prior art analysis maps onto these phases differently.<\/p>\n\n\n\n

During initial screening, the appropriate AI tool is portfolio mapping and PTAB risk scoring \u2014 using DrugPatentWatch data and pre-trained PTAB prediction models to generate a rapid probabilistic assessment of the top five to ten patents. This takes two to three days and produces a go\/no-go signal on IP risk materiality before significant deal resources are committed.<\/p>\n\n\n\n

During confirmatory due diligence, full AI prior art search and claim charting should run on the load-bearing patents identified during screening. This is where the external platform engagement and outside counsel review occur. The output \u2014 abbreviated invalidity opinions with quantitative prior art strength estimates \u2014 feeds the deal model before the LOI economics are locked in.<\/p>\n\n\n\n

During final negotiation, the IP risk analysis shifts from probabilistic to contractual. Where material prior art vulnerability has been identified, the acquirer should seek representations and warranties from the target on the completeness of prior art disclosure to the USPTO, specific indemnification provisions tied to the identified vulnerabilities, holdbacks or escrow provisions sized to the expected value loss in the scenarios where the vulnerable patents are invalidated, and MAC clauses that specifically define material adverse change to include PTAB institution decisions on the key patents.<\/p>\n\n\n\n

The Valuation Model Structure<\/strong><\/h3>\n\n\n\n

A pharmaceutical deal model that incorporates AI prior art analysis should treat each key patent as a separate scenario node in the DCF. The base case assumes full patent protection through the nominal expiry date. The downside case assumes invalidation of each key patent at a probability estimated from the AI prior art analysis output. The weighted average of these scenarios \u2014 the “AI-adjusted NPV” \u2014 replaces the binary patent expiry assumption in the traditional model.<\/p>\n\n\n\n

For a target with three load-bearing patents, this produces a matrix of eight scenarios (each patent valid or invalid), weighted by the probability estimates from the PTAB risk model. The weighted average scenario produces an IP-risk-adjusted deal value that can be compared directly to the offer price. If the offer price exceeds the IP-risk-adjusted NPV by a margin that exceeds the deal team’s risk tolerance, the gap needs to be addressed either through price reduction or contractual protection.<\/p>\n\n\n\n

\n\n\n\n

What Changes When the Acquirer Is Also the IPR Petitioner<\/strong><\/h2>\n\n\n\n

One strategic scenario that AI prior art analysis enables, which traditional due diligence does not contemplate, is the acquirer using the identified prior art not for defensive purposes but as an offensive tool in pre-acquisition negotiations or competitive interactions.<\/p>\n\n\n\n

An acquirer who identifies strong prior art against a potential target’s key patent has a distinct option that neither party to the transaction can easily exploit post-closing: the option to file an IPR petition before the acquisition, using the prior art to reduce the target’s patent valuation before committing to a price. This is legally permissible \u2014 any party can file an IPR petition \u2014 but operationally complex because it reveals the acquirer’s interest in the target and creates estoppel consequences that would affect the acquirer’s position post-closing.<\/p>\n\n\n\n

More practically, the identified prior art can be used in price negotiations directly. An acquirer who presents a target with a credible invalidity analysis \u2014 here is the prior art we found, here is the claim chart showing how it maps to your composition of matter patent, here is our probability estimate for PTAB institution \u2014 negotiates from a different position than an acquirer relying solely on the target’s representations about its patent strength. The prior art analysis converts a qualitative debate about IP strength into a quantitative exchange that anchors the negotiation in evidence rather than assertion.<\/p>\n\n\n\n

\n\n\n\n

The Regulatory and Ethical Dimension<\/strong><\/h2>\n\n\n\n

AI prior art analysis raises one regulatory issue and one ethical issue that practitioners should be aware of.<\/p>\n\n\n\n

The regulatory issue concerns AI-generated prior art disclosures. Under 37 C.F.R. \u00a7 1.56, patent applicants have a duty of candor to the USPTO, which includes disclosure of known material prior art. If an AI prior art search conducted during M&A due diligence surfaces documents that are material to the patentability of claims in pending applications within the acquired company’s portfolio, those documents may need to be disclosed to the USPTO through an Information Disclosure Statement post-closing. Failure to disclose material prior art of which the patent owner became aware can support an inequitable conduct defense in future litigation. Post-closing patent housekeeping should include a review of the AI prior art search outputs against pending applications.<\/p>\n\n\n\n

The ethical issue concerns the use of AI-generated invalidity arguments. The argument that patent thickets harm patients by extending monopoly pricing has been made credibly in the Humira context, the Soliris context, and numerous others. AI tools that make it more efficient to identify and invalidate weak secondary patents serve a public interest function, not merely a commercial one. Acquirers who use AI prior art analysis to understand what they are actually buying \u2014 and to avoid overpaying for patent portfolios whose exclusivity value is built on patents that should not have been granted \u2014 are participating in a market correction that ultimately benefits both innovation efficiency and patient access.<\/p>\n\n\n\n

\n\n\n\n

The Next Generation: Multimodal AI and Agentic Search<\/strong><\/h2>\n\n\n\n

The platforms described in this article represent the current state of AI prior art analysis. The direction of travel is toward capabilities that are meaningfully more powerful for pharmaceutical patents specifically.<\/p>\n\n\n\n

Multimodal chemical AI \u2014 systems that can natively process protein structures in PDB format, chemical structures in SMILES notation, 3D molecular models, and spectroscopic data alongside patent text \u2014 will enable prior art searches that cover the full representation space of pharmaceutical innovation [4<\/a>]. A prior art search that is simultaneously semantic over text and structural over molecular graphs can identify anticipating references that contain the relevant structure but describe it using entirely different language than the patent claims. For biologic patents where the claim scope is defined partly by structural characteristics of the antibody, this multimodal capability is qualitatively different from text-only semantic search.<\/p>\n\n\n\n

Agentic patent systems \u2014 AI systems given goal-level instructions rather than single queries \u2014 are beginning to emerge as a distinct category [4<\/a>]. In an agentic framework, the instruction is not “find prior art for claim element X” but “determine whether this patent is invalidable and generate the strongest available invalidity argument.” The agent executes the full search-analyze-synthesize cycle autonomously, returning an invalidity opinion rather than a set of search results. For M&A due diligence under time pressure, this capability would compress the AI prior art analysis cycle further \u2014 potentially from days to hours for the initial retrieval and charting stages.<\/p>\n\n\n\n

Merck’s 2025 acquisition of Atomwise for $2.1 billion included specific contractual provisions requiring human oversight of all generative chemistry outputs, a reflection of major pharmaceutical companies’ awareness that AI-generated scientific claims create patentability risks [1<\/a>]. The reverse of that risk \u2014 using AI to audit the patentability of acquired IP \u2014 is the capability this article has described. Both applications will mature as the underlying AI systems improve.<\/p>\n\n\n\n

\n\n\n\n

Key Takeaways<\/strong><\/h2>\n\n\n\n
    \n
  • Patent valuation in pharmaceutical M&A is not binary.<\/strong> Effective patent exclusivity is a probability distribution over time, not a hard date. AI prior art analysis enables quantitative modeling of that distribution by estimating PTAB invalidation probability from the prior art evidence quality.<\/li>\n\n\n\n
  • The prior art most likely to kill a pharmaceutical patent is the prior art least likely to be found by keyword search.<\/strong> Non-patent literature \u2014 conference proceedings, regulatory documents, foreign-language academic papers, technical bulletins \u2014 is systematically underrepresented in traditional due diligence and well-suited to AI semantic search.<\/li>\n\n\n\n
  • PTAB statistics should be in every pharmaceutical deal memo.<\/strong> In the first half of 2024, 71% of PTAB trials resulted in all challenged claims being cancelled. That number applies to patents you inherit at closing unless the prior art analysis shows otherwise.<\/li>\n\n\n\n
  • Load-bearing patents need full AI treatment.<\/strong> Not every patent in a target’s portfolio warrants deep AI prior art analysis. Triage using DrugPatentWatch portfolio mapping to identify the two to four patents that actually underpin the exclusivity value, then conduct full semantic search and claim charting on those patents specifically.<\/li>\n\n\n\n
  • AI retrieval and attorney review are complements, not substitutes.<\/strong> AI prior art platforms provide coverage and speed; qualified patent counsel provides claim construction accuracy and legal opinion. The optimal workflow uses both, with AI on retrieval and organization, attorney on construction and opinion.<\/li>\n\n\n\n
  • Post-closing vulnerability mapping is equally important.<\/strong> The acquirer who maps patent vulnerabilities before competitors file IPR petitions has strategic options \u2014 continuation applications, licensing adjustments, litigation positioning \u2014 that the acquirer who discovers vulnerabilities in an institution decision does not.<\/li>\n\n\n\n
  • The UPC adds a new global risk dimension.<\/strong> For European market revenues, a single UPC revocation action can eliminate patent protection across participating EU member states. AI prior art analysis for global pharmaceutical M&A should explicitly address UPC vulnerability as a distinct scenario.<\/li>\n<\/ul>\n\n\n\n
    \n\n\n\n

    Frequently Asked Questions<\/strong><\/h2>\n\n\n\n

    1. How much does an AI-powered prior art search actually reduce due diligence time compared to traditional methods?<\/strong><\/h3>\n\n\n\n

    For a full prior art search across a pharmaceutical patent portfolio of ten to fifteen load-bearing patents, a traditional keyword-based search with human review requires four to six weeks of dedicated effort by a team of three to four patent professionals. An AI-augmented workflow using platforms like Patlytics or DeepIP, configured with pharmaceutical-specific search corpora, compresses the retrieval and initial claim charting stages to three to five business days. Attorney review of the AI-generated outputs adds one to two weeks depending on team size and patent complexity. Total elapsed time: two to three weeks, versus four to six. More important than the time reduction is the coverage improvement: the AI workflow accesses non-patent literature and cross-lingual references that the traditional keyword workflow would miss entirely, regardless of how much time was allocated.<\/p>\n\n\n\n

    2. Can AI prior art analysis identify prior art that the original patent examiner missed but that the USPTO examiner was responsible for finding?<\/strong><\/h3>\n\n\n\n

    Yes, and this is one of the most important use cases. USPTO examiners are under time constraints \u2014 examiners average approximately 20-25 hours per application across all examination stages, including prior art search. That time allocation is insufficient for comprehensive non-patent literature search across the full corpus of biomedical, chemical, and pharmaceutical sciences literature. AI semantic search, which can process millions of documents in minutes, is not subject to those time constraints. The result is that AI prior art searches routinely identify references that were publicly available before the patent filing date and that the examiner did not cite, but that a PTAB petitioner or district court defendant could use to challenge the patent’s validity. The fact that the examiner did not find a reference does not make it legally irrelevant \u2014 the PTAB will consider prior art that was not before the examiner during prosecution, and that art will be weighed under the preponderance of evidence standard.<\/p>\n\n\n\n

    3. What is the legal status of an AI-generated claim chart in an IPR proceeding?<\/strong><\/h3>\n\n\n\n

    An AI-generated claim chart is not itself a legal filing \u2014 it is an analytical document that must be reviewed, adopted, and signed by a registered patent practitioner before it can be submitted in an IPR proceeding. The PTAB requires that IPR petitions be prepared by and bear the signature of a registered practitioner. However, the substance of the chart \u2014 the prior art references identified and the mapping of those references to specific claim limitations \u2014 is legally significant regardless of how it was generated. If an AI system identifies a prior art reference and generates a technically accurate claim chart mapping that reference to the patent claims, the chart remains valid as evidence if an attorney reviews it, confirms the technical accuracy, and adopts it as the basis for the invalidity argument in the petition. Several IPR petitions filed in 2024 and 2025 have used AI-generated preliminary claim charts as the starting point for attorney-refined petitions, and the PTAB has not raised objections to the process \u2014 only to the content.<\/p>\n\n\n\n

    4. How should a pharmaceutical acquirer handle a situation where AI due diligence identifies strong invalidating prior art post-closing?<\/strong><\/h3>\n\n\n\n

    The immediate priorities are two: legal disclosure assessment and strategic positioning. On disclosure: if any of the identified prior art documents are material to pending patent applications within the acquired portfolio, counsel should assess whether an Information Disclosure Statement is required to comply with the duty of candor under 37 C.F.R. \u00a7 1.56. Failure to disclose material prior art that the patent owner knows about can constitute inequitable conduct. On strategic positioning: the acquired company should assess whether the vulnerable claims can be addressed through continuation practice \u2014 filing continuation applications with narrowed claims that design around the prior art \u2014 and whether any pending litigation or licensing programs should be adjusted to reflect the identified vulnerability. Contractually, if the acquisition agreement included standard IP representations and warranties, an acquirer who discovers material undisclosed prior art post-closing should evaluate whether that constitutes a breach of the representation. Under most acquisition agreement structures, the acquirer would need to demonstrate that the prior art was known to the target at signing and was not disclosed, which is factually difficult to establish for prior art that was genuinely not found during the target’s own prosecution or due diligence process.<\/p>\n\n\n\n

    5. Are AI prior art search platforms reliable enough for major pharmaceutical transactions, or are they still in an early, experimental stage?<\/strong><\/h3>\n\n\n\n

    The leading platforms \u2014 Patlytics, DeepIP, IPRally, and PatentScan \u2014 have been used in production legal and M&A due diligence contexts since 2022-2023 and have a sufficient track record to assess their reliability. The more accurate characterization is that they are reliable for their designed function \u2014 retrieval and initial claim charting \u2014 but should not be used as replacements for attorney claim construction and legal opinion. Evaluation studies comparing AI platform outputs against the prior art eventually surfaced in IPR proceedings have generally found that the platforms identify 60-80% of the ultimately relevant prior art references, with the gap explained primarily by very narrow NPL categories (specific conference proceedings from niche scientific fields) that are not yet in the platforms’ corpora. For M&A due diligence purposes, 60-80% coverage of the relevant prior art is vastly better than the 10-20% coverage typical of traditional keyword due diligence \u2014 particularly given that the AI-assisted coverage is concentrated in the NPL categories that are most likely to contain the strongest invalidating references. The platforms are not experimental. They are production tools with known limitations that an informed user can account for.<\/p>\n\n\n\n

    \n\n\n\n

    References<\/strong><\/h2>\n\n\n\n
      \n
    1. AstraZeneca PLC. (2022, March 17). AstraZeneca reaches settlement agreement resolving patent litigation related to Ultomiris<\/em>. AstraZeneca Media Centre. https:\/\/www.astrazeneca.com\/media-centre\/press-releases\/2022\/astrazeneca-reaches-settlement-agreement-resolving-patent-litigation-related-to-ultomiris.html<\/li>\n\n\n\n
    2. Fierce Pharma. (2025, April 17). AstraZeneca’s Alexion accused of extending Soliris monopoly through sham patents in new suit<\/em>. https:\/\/www.fiercepharma.com\/pharma\/astrazenecas-alexion-accused-extending-soliris-monopoly-through-sham-patents-new-suit<\/li>\n\n\n\n
    3. Finnegan. (2024). Trends in PTAB trials involving drug and biologic patents<\/em>. Finnegan, Henderson, Farabow, Garrett & Dunner, LLP. https:\/\/www.finnegan.com\/en\/insights\/blogs\/at-the-ptab-blog\/trends-in-ptab-trials-involving-drug-and-biologic-patents.html<\/li>\n\n\n\n
    4. DrugPatentWatch. (2026, March 12). The future of patent intelligence tools: How AI is revolutionizing the landscape<\/em>. https:\/\/www.drugpatentwatch.com\/blog\/the-future-of-patent-intelligence-tools-how-ai-is-revolutionizing-the-landscape\/<\/li>\n\n\n\n
    5. Cypris. (2025). Best prior art search automation tools in 2025<\/em>. https:\/\/www.cypris.ai\/insights\/best-prior-art-search-automation-tools-in-2025<\/li>\n\n\n\n
    6. DeepIP. (2025). Invalidity search: AI-powered prior art discovery<\/em>. https:\/\/www.deepip.ai\/products\/invalidity-search<\/li>\n\n\n\n
    7. USPTO Patent Trial and Appeal Board. (2024). PTAB statistics: FY2024 end of year outcome roundup<\/em>. United States Patent and Trademark Office.<\/li>\n\n\n\n
    8. DrugPatentWatch. (2026, March 24). The Patent Trial and Appeal Board: The definitive analyst’s guide to IPR strategy, pharmaceutical IP valuation, discretionary denial mechanics, and the post-Arthrex power shift<\/em>. https:\/\/www.drugpatentwatch.com\/blog\/understanding-the-patent-trial-and-appeal-board-ptab-a-comprehensive-overview\/<\/li>\n\n\n\n
    9. TT Consultants. (2025, March 19). Patent invalidation trends: PTAB impact & global developments<\/em>. https:\/\/ttconsultants.com\/the-global-shift-in-patent-invalidation-ptab-upc-emerging-trends\/<\/li>\n\n\n\n
    10. DrugPatentWatch. (2025, November 19). How safe is your drug patent from PTAB challenges? A strategic guide for pharma leaders<\/em>. https:\/\/www.drugpatentwatch.com\/blog\/how-safe-is-your-drug-patent-from-ptab-challenges-a-strategic-guide-for-pharma-leaders\/<\/li>\n\n\n\n
    11. DrugPatentWatch. (2026). AstraZeneca branded and generic drugs, international patents<\/em>. https:\/\/www.drugpatentwatch.com\/p\/applicant\/Astrazeneca<\/li>\n\n\n\n
    12. DrugPatentWatch. (2026, March 22). Drug patent strategy: The definitive guide for pharmaceutical IP teams, R&D leads, and institutional investors<\/em>. https:\/\/www.drugpatentwatch.com\/blog\/optimizing-your-drug-patent-strategy-a-comprehensive-guide-for-pharmaceutical-companies\/<\/li>\n\n\n\n
    13. Ropes & Gray LLP. (2024, August). The transformative impact of AI on patent prior art searches<\/em>. https:\/\/www.ropesgray.com\/en\/insights\/alerts\/2024\/08\/the-transformative-impact-of-ai-on-patent-prior-art-searches<\/li>\n\n\n\n
    14. TT Consultants. (2025, December 23). Patent invalidation in 2025: Key grounds, legal frameworks & recent trends<\/em>. https:\/\/ttconsultants.com\/patent-invalidation-in-2025-key-grounds-legal-frameworks-recent-trends\/<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"

      Every pharmaceutical acquisition has a number at its core that almost nobody states plainly: the price you are paying for […]<\/p>\n","protected":false},"author":1,"featured_media":38736,"comment_status":"open","ping_status":"closed","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-38733","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\/38733","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=38733"}],"version-history":[{"count":1,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/posts\/38733\/revisions"}],"predecessor-version":[{"id":38737,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/posts\/38733\/revisions\/38737"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/media\/38736"}],"wp:attachment":[{"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/media?parent=38733"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/categories?post=38733"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/tags?post=38733"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}