Artificial intelligence is rapidly transforming drug discovery by generating candidate molecules, predicting biological activity, and optimizing pharmacological properties with minimal human intervention. These advances raise a fundamental patent law question: are inventions arising from AI-assisted drug discovery patent-eligible under 35 U.S.C. §101, or do they fall into judicial exceptions for abstract ideas, natural phenomena, or laws of nature?
The answer depends on how courts characterize both the role of the algorithm and the nature of the claimed invention.
The §101 Framework
Under Supreme Court precedent, patent eligibility requires more than reciting an abstract idea or natural law implemented on a computer. Courts apply a two-step test: first, determine whether the claim is directed to a judicial exception; second, assess whether the claim contains an “inventive concept” sufficient to transform the exception into patent-eligible subject matter.
In biotechnology, this framework has been shaped by decisions excluding naturally occurring DNA and invalidating diagnostic method claims that merely apply natural correlations. In software, claims directed to data processing or mathematical algorithms without technological improvement have similarly failed.
AI-assisted drug discovery implicates both domains: algorithms and biological relationships.
Claims to Computational Methods
Method claims directed to using a machine learning model to predict molecular properties or generate chemical structures are vulnerable to characterization as abstract ideas. If framed as mathematical optimization or pattern recognition performed on generic computers, such claims risk invalidation under the Alice line of cases.
However, courts have upheld software claims that improve computer functionality or solve a technological problem in a specific way. Claims that recite concrete model architectures, training regimes, or data representations tailored to molecular design may be more likely to survive.
The key distinction is between claiming the result—“use AI to identify a drug candidate”—and claiming a specific technical process for doing so.
Claims to Molecules Generated by AI
A different analysis applies when the claim is directed to the molecule itself rather than to the algorithm that produced it. A synthetic compound or engineered protein does not become ineligible merely because it was designed computationally.
Patent eligibility turns on whether the claimed molecule is a product of nature or a human-made composition of matter. AI-designed molecules with no natural analogs are unlikely to be excluded on natural phenomenon grounds. They resemble traditional chemical inventions in which design tools were used to arrive at a structure.
Thus, while method claims to AI-driven discovery may face §101 hurdles, composition claims to AI-designed drugs may avoid them entirely.
Diagnostic and Correlation-Based Risks
Problems arise where AI is used to discover correlations between molecular structures and biological effects, and the claim is framed around that relationship. For example, a claim stating that a particular molecular pattern correlates with therapeutic efficacy risks characterization as a law of nature if it merely reports a discovered relationship.
This echoes the diagnostic patent cases, where identifying a biological correlation and instructing others to apply it has been held ineligible. If AI reveals that certain chemical features predict binding to a target, claiming that insight alone may not suffice.
To avoid this trap, applicants must anchor claims in concrete applications: specific molecules, specific manufacturing methods, or specific treatment regimens.
The “Inventive Concept” in AI Drug Discovery
Even if a claim is directed to an abstract idea or natural law, it may still be patent-eligible if it contains an inventive concept. In AI-assisted drug discovery, possible sources of inventiveness include:
- A novel neural network architecture tailored to chemical space,
- A specialized training dataset or feature representation,
- Integration of AI outputs with experimental validation steps,
- A new chemical class discovered through algorithmic design.
Claims that merely automate known drug discovery steps with generic AI tools are unlikely to satisfy this requirement. By contrast, claims that solve a technical problem in molecular design using a specific, unconventional approach may be eligible.
Convergence of §101 and §112
In practice, §101 challenges to AI-assisted drug discovery claims may blur into enablement and written description issues. If a claim broadly recites “using AI to generate a therapeutic compound” without specifying how, it risks invalidation as both abstract and insufficiently disclosed.
Courts may increasingly treat vague AI claims as ineligible because they fail to describe a concrete technological application. Thus, eligibility may depend as much on disclosure quality as on claim category.
Policy Considerations
From a policy perspective, excluding AI-assisted drug discovery from patent eligibility could undermine incentives for investment in computational biology. Conversely, allowing overly broad claims to AI-based discovery methods could preempt entire fields of research.
The judicial exceptions doctrine reflects a concern with monopolizing basic tools of science. AI models trained on biological data may be viewed as new tools of discovery. Granting patents on the use of such tools in general terms risks foreclosing downstream innovation.
This tension suggests that courts will likely permit patents on specific AI-designed molecules and narrowly tailored algorithms, while rejecting claims that attempt to capture AI-driven discovery at a conceptual level.
Likely Trajectory
In the near term, AI-assisted drug discovery will likely be patent-eligible when:
- The claim is directed to a non-natural molecule or composition of matter.
- The method claim recites a specific technical process, not a generic application of AI.
- The invention involves more than discovering a correlation between molecular features and biological function.
Conversely, claims are likely to fail where they:
- Recite AI at a high level of abstraction,
- Claim biological relationships without concrete application, or
- Seek to monopolize the use of AI in drug discovery broadly.
Conclusion
AI-assisted drug discovery sits at the intersection of software and biotechnology patent law, invoking concerns about abstract ideas, natural laws, and technological innovation. While the use of AI does not, by itself, render an invention ineligible, the manner in which claims are framed will be decisive. The lesson is clear: anchor AI-driven discoveries in tangible chemical inventions or in specific technological improvements. In an era where algorithms increasingly shape molecular design, patent eligibility will turn not on the presence of AI, but on whether the claimed invention represents a concrete advance in chemistry rather than a computational insight into nature.
BioTech IP Review 
Leave a comment