AI-Generated Data, Patent Protection, And The Implications For Clinical Supply

By Austin J. Kim, Senior Counsel and Intellectual Property Lawyer, Foley & Lardner LLP

Artificial intelligence (AI)-powered tools are already accelerating biomedical R&D in ways that directly impact development timelines and clinical trial supply planning, including anticipating patient enrollment, dosing schedules, and inventory needs. DeepMind’s AlphaFold can in a matter of minutes provide accurate predictions of protein structures, a task that once required months or years of experimental work and specialized equipment. Generative platforms such as ChatGPT can process vast repositories of data and produce human-like responses in seconds. These capabilities are particularly compelling in areas where gathering traditional in vitro or in vivo clinical or experimental data sets can be prohibitively costly or even impossible, such as with rare diseases with very small patient populations.AI Integration In Diagnostics, Therapeutics, And Clinical Supply

The biomedical industry is actively exploring how to embed AI tools in all stages of developing diagnostics (e.g., biomarker assays, imaging methods, or medical devices) and therapeutics (e.g., drugs, biologics, cell and gene therapies, or dosing regimens). For clinical supply teams, these AI-driven insights can inform trial design, inventory forecasting, and cold chain logistics, helping ensure the right drug reaches the right patient at the right time. Stakeholders should consider how the integration of AI tools in R&D impacts their intellectual property (IP) strategy, and vice versa. 

Patenting diagnostic and therapeutic innovations has always posed special challenges. Diagnostic inventions have to show how the invention is non-conventional compared to previous techniques, not just a bare correlation or abstract idea. With therapeutic inventions, there has to be a credible demonstration of the prophylactic effect. The use of generative AI raises new questions in patenting inventions in the biomedical field, which can also intersect with clinical trial supply considerations, such as demonstrating that a therapy can be reliably produced and delivered at scale for trial populations. 

AI-Generated Data And Regulatory Insights

The U.S. FDA is already using AI-generated data to evaluate safety and efficacy of drugs and medical devices.¹ While these AI tools inform R&D, they can also provide insights critical to clinical trial supply, such as predicting drug demand, optimizing inventory, and anticipating cold chain requirements. While the United States Patent and Trademark Office (USPTO) has commented on the patentability of AI inventions² and directed that AI agents cannot be named as inventors,³ it has yet to wade in on the acceptability of AI-generated in silico data. 

There are several types of AI‑generated data that can be relevant to diagnostic and therapeutic inventions. This section highlights three examples: 

• Structure/Sequence Outputs: Tools such as AlphaFold can be used to generate detailed molecular structures, potential binding sites, and functional predictions.  These outputs can be used to prioritize candidate molecules before costly in vitro or in vivo work. They can also help clinical supply teams plan production batches and anticipate material needs before in vitro or in vivo work. For clinical supply teams, the value is not simply earlier visibility into potential candidates but the ability to translate that information into earlier manufacturing and supply decisions. Earlier identification of likely candidates can influence material requirements, batch planning, lead times, and capacity commitments before clinical demand is fully known.

• Simulations: Digital twins can simulate patient-specific disease states or treatment responses, and physiologically based pharmacokinetic (PBPK) models can forecast how a drug behaves in the body. These data can enable virtual trials and provide clinical supply teams with early insights on dosing schedules, storage, and distribution logistics. As these models refine assumptions about patient populations, dosing, and treatment response, clinical supply teams can use those signals to revisit demand assumptions before production and distribution decisions become difficult or costly to change. The earlier those assumptions can be incorporated into supply planning, the more flexibility teams have to adjust inventory and manufacturing strategies as the trial evolves.

• Generative Outputs: Generative models, such as large language models (LLMs) and generative adversarial networks (GANs), can produce data sets modeled on actual patient data or extract relevant signals from electronic health records. Such outputs can support supply planning by predicting patient population needs and potential variability in trial demand.

Patient Eligibility, Enablement, And Technical Improvements

For diagnostic and therapeutic inventions, the ability to secure patent protection is dependent on data. To obtain patent protection, an invention must be directed to patentable subject matter, be novel and non-obvious, and meet the written description and enablement requirements. Data generated through AI can also help clinical supply teams plan and justify trial logistics by predicting material needs, dosing vulnerability, and supply risks.  

One of the largest hurdles to patenting new diagnostic inventions is patent subject matter eligibility under 35 U.S.C. § 101. 

Under this requirement, inventions that are deemed as an abstract idea, laws of nature, or natural phenomena are excluded from eligibility.⁴ Under the USPTO’s examination guidelines, when an invention is deemed to fall under one of these excluded categories, examiners are to evaluate whether the claims integrate the exception into a practical application (e.g., by providing a technical improvement to an existing technology).⁵ If there is no integration, examiners are to assess whether the claimed features amount to something “significantly more,” such as by showing that the features are not well understood, routine, or conventional.

Diagnostic inventions are often challenged as directed to one of these categories of exclusion.⁶ With the increasing incorporation of AI, diagnostic inventions are often an amalgamation of AI models with biomedical data to provide output used for clinical applications, which can also inform clinical supply decisions, such as anticipating patient enrollment patterns, planning drug distribution, and optimizing inventory allocation for trial sites. 

The USPTO’s 2024 guidance on AI inventions provided a specific example relating to the use of an AI model in personalized medical treatment.⁷ In this example, a claim that merely stated “administering an appropriate treatment” was rejected because the claim did not specify what treatment for which patient population.⁸ In contrast, a claim that specified the particular treatment for a specific patient population was found eligible. The incorporation of an administration step with a specific treatment to a specific patient population can help to resolve the patent eligibility issues. Adding a step of administering treatment in a patent claim, however, may not be desirable due to the issue of split infringement: one actor could be running the diagnostic, while another actor could be performing the administration. This also highlights clinical supply challenges, where different teams may be responsible for manufacturing, distribution, and site administration, making coordination and logistics critical.

There are pathways forward.  Notably, in one Federal Circuit case, a diagnostic patent was upheld partly because the patent in question documented clear technical improvements, namely reducing false positives and negatives in cardiac monitoring.⁹ AI-generated data could be potentially used to show that the invention provides a technical improvement and is non-conventional relative to the prior techniques. For clinical supply teams, using virtual patient populations or simulations can also help plan trial inventory, anticipate drug demand, and ensure adequate supply for high-risk populations. Applicants should collaborate with inventors to articulate how their inventions achieve these improvements and identify which steps are non-conventional. 

Therapeutics, Enablement, And Clinical Supply Implications

Another major challenge in patenting therapeutics-related inventions is in calibrating the scope of the claims with the written description and enablement requirements under 35 U.S.C. § 112. Under this section, a patent application must describe the invention with adequate detail such that a person skilled in the art can recognize the boundaries of the invention (i.e., the “written description” requirement) and can be enabled to practice the invention (i.e., the “enablement” requirement).¹⁰ This level of detail can also help clinical supply teams understand dosing requirements, batch sizes, and potential logistical complexities for trial implementation.

Therapeutics often operate within complex, unpredictable biological systems. This unpredictability means that any claims about therapeutic effect are closely scrutinized. Without credible supporting evidence, patents risk being rejected or forced into narrower claim scopes. This same unpredictability can create significant clinical trial supply risk, including over- or underestimating drug demand, misaligning batch production, or struggling with dose adjustment mid-trial.

AI-generated data can give early-stage teams more options to bolster their applications. For example, if there is already experimental data demonstrating therapeutic activity for a few antibody or small molecule examples, AI‑generated data could be used to support broader genus claims by showing numerous related variants that are predicted to have similar activity.  In parallel, these insights can help clinical supply teams anticipate formulation needs, stability considerations, and scalability for multiple variants before large-scale manufacturing begins.

In another example, data from PBPK simulations could lend support to a credible therapeutic effect of a drug or a medical device. These simulations can also inform dosing strategies, packaging configurations, and distribution models for global site trials. The patent application should include an explanation of the reliability of the model and data and include descriptions of the training method and validation cohorts. 

Prophecy, Provisional Filings, And Clinical Trial Supply

Since the USPTO has yet to comment on the acceptability of in silico data, applicants should take a belt-and-suspenders approach by also including prophetic examples to accompany preliminary AI-generated data. Prophetic examples describe expected future or anticipated results, without actual past experimental data.¹¹ Should there be pushback during prosecution on the use of AI-generated data, post-filing submissions of clinical data that fit within the scope of the prophetic examples could be used to support the claimed therapeutic effect. This staged approach can also align with clinical supply timelines, allowing teams to gather real-world trial data while refining forecasts, inventory strategies, and site distribution plans. It should be noted, however, that the acceptance of prophetic examples and post-filing submissions varies by jurisdiction, with many jurisdictions outside the U.S. and Europe not accepting them as evidence of enablement.

To provide additional time to gather clinical data, a provisional application can be a smart tool. This type of application can lock in an early priority date while giving teams up to 12 months to refine the invention, accumulate more clinical data, and gain clarity as to the commercial viability of the diagnostic or therapeutic invention. This window can also give clinical supply leaders time to align manufacturing capacity, qualify CMOs, and build a more resilient trial supply strategy before full commitment.

AI, IP, And The Future Of Clinical Trial Supply

AI‑generated data is becoming a powerful asset in biomedical innovation. Leveraging in silico data alongside traditional in vivo or in vitro data can strengthen patent filings for diagnostics and therapeutics. When strategically integrated, these same data streams can reduce clinical trial supply uncertainty, improve forecast accuracy, and support more efficient global trial execution. While the USPTO has yet to comment on this issue, the ever-growing use of AI makes it likely that this issue will be faced sometime in the future. Stakeholders, including clinical supply leaders, should continue to monitor legal developments in this space. The operational question for clinical supply leaders is therefore not simply whether AI-generated data is accurate but when that data becomes reliable enough to influence supply decisions. Using those signals too early can introduce new uncertainty, while waiting too long can leave teams with less time to adjust manufacturing, inventory, and distribution plans.

References

About The Author:

Austin Kim is a senior counsel and intellectual property lawyer with Foley & Lardner LLP. He is a member of the firm’s Electronics Practice as well as a member of the Automotive Industry Team and Innovative Technology Sector. Austin counsels clients in strategic patent portfolio development through various stages in procurement and across multiple jurisdictions. He has prepared and prosecuted applications in the electric vehicle space, including fuel cells, autonomous driving, and remote controls. He also has experience in telecommunications, virtualization, cloud computing, web applications, and biomedical devices.