BioTech IP Review

As artificial intelligence becomes an increasingly central tool in molecular design, a new question emerges for patent law: can a molecule generated by an AI system be deemed “obvious” under 35 U.S.C. §103? Traditional obviousness doctrine assumes human-driven experimentation guided by known chemical principles. AI-driven molecular generation, however, operates by probabilistic search across immense chemical spaces, challenging established conceptions of what it means for an invention to be predictable or routine.

The Classical Obviousness Framework

Under §103, a patent claim is invalid if the differences between the claimed invention and the prior art would have been obvious to a person having ordinary skill in the art (PHOSITA) at the time of invention. In chemistry and biotechnology, courts often look to:

• Structural similarity to known compounds,
• Reasonable expectation of success, and
• Motivation to modify prior art molecules.

For small molecules, obviousness frequently turns on whether a skilled chemist would have been motivated to modify a known lead compound in a particular way with a reasonable expectation of retaining or improving biological activity. In protein and biologic contexts, the analysis often centers on sequence homology and functional predictability.

This framework presumes that invention proceeds through rational human reasoning informed by known structure-function relationships.

AI as a New “Ordinary Skill” Baseline

AI-generated molecules destabilize this presumption. Modern generative models can propose candidate structures by sampling high-dimensional chemical space using learned statistical patterns rather than explicit chemical heuristics. These systems may output structures that do not resemble known leads and were not suggested by conventional medicinal chemistry logic.

The question thus becomes whether the relevant benchmark for obviousness should remain the unaided human chemist or should incorporate algorithmic design tools as part of the PHOSITA’s skill set.

Courts have long recognized that obviousness is assessed in light of the tools available to the PHOSITA. Just as computer modeling and high-throughput screening altered expectations of predictability in drug discovery, AI-based generative models may similarly expand what is considered “routine optimization.”

If AI becomes a standard tool in molecular design, then molecules produced by such systems may be deemed obvious if they arise from routine application of known algorithms to known targets.

Predictability Versus Search

A core tension arises between predictability and search. AI systems often succeed not because they “understand” chemistry but because they can explore vast combinatorial spaces efficiently. Their outputs may be statistically optimized but mechanistically opaque.

In classical doctrine, unpredictability favors nonobviousness. If the relationship between structure and function is uncertain, even small molecular changes may support patentability. AI, however, can overcome unpredictability through brute-force probabilistic exploration.

This raises a doctrinal dilemma: does computational success convert an unpredictable field into a predictable one for §103 purposes?

If predictability is defined by outcome rather than reasoning, courts may conclude that AI renders certain classes of molecules obvious because they can be routinely generated given sufficient data and compute power. Conversely, if predictability is defined by mechanistic insight, AI outputs may remain nonobvious because their functional success could not have been anticipated by human reasoning.

Structural Similarity and Prior Art

Another axis of analysis is structural similarity to known compounds. AI-generated molecules are often optimized against known targets using datasets that include existing ligands. As a result, outputs may share substructures with prior art compounds even if the overall scaffold is novel.

Under current law, close structural similarity coupled with known activity may create a prima facie case of obviousness. Applicants must then rebut with evidence of unexpected properties.

However, if an AI-generated molecule is structurally dissimilar to known ligands but functionally effective, the usual motivation-to-modify logic weakens. No human chemist would necessarily have been motivated to construct that structure from known compounds.

Thus, AI outputs may bifurcate into two categories: those that resemble prior art and are vulnerable to obviousness challenges, and those that are architecturally novel and resistant to conventional §103 reasoning.

The Role of “Routine” Experimentation

Obviousness doctrine frequently turns on whether an invention was achieved through routine experimentation. AI systems automate what was once iterative medicinal chemistry, potentially transforming exploratory research into a push-button exercise.

Patent challengers may argue that once a target and training dataset are known, running a generative model constitutes routine optimization, rendering the resulting molecules obvious.

Applicants, in contrast, will argue that the invention lies not in running the model but in identifying a molecule with the claimed biological properties, particularly where only a small fraction of generated candidates succeed experimentally.

Courts may thus confront whether algorithmic trial-and-error qualifies as routine experimentation or whether it remains inventive due to the stochastic nature of model outputs.

Secondary Considerations

Objective indicia of nonobviousness may take on renewed importance in AI-generated molecule cases. Evidence of:

• Unexpected potency or selectivity,
• Improved safety or pharmacokinetics,
• Long-felt unmet need, or
• Commercial success,

may help counter claims that AI merely automated what would have been discovered anyway.

However, such arguments may be weakened if AI is shown to systematically outperform human chemists in generating effective molecules, suggesting that success is an expected result of applying the algorithm.

A Shifting Legal Standard

As AI tools become embedded in drug discovery workflows, courts may implicitly redefine the PHOSITA as someone equipped with generative models and massive datasets. Under such a standard, molecules that emerge from straightforward application of known AI methods to known biological targets may increasingly be deemed obvious.

At the same time, molecules that result from novel model architectures, unconventional training regimes, or unanticipated functional behaviors may retain nonobviousness due to their departure from established computational practices.

Conclusion

AI-generated molecules expose a fault line in §103 doctrine between human reasoning and machine search. If obviousness is grounded in what a skilled human would predict, many AI-designed compounds may remain nonobvious. If obviousness is grounded in what a skilled system can routinely generate, the bar for patentability will rise. This emerging issue will require careful framing of invention narratives. The focus may shift from the molecule itself to the technical obstacles overcome in generating and validating it. In the age of algorithmic chemistry, non-obviousness may depend less on molecular structure and more on the ingenuity embedded in the system that produced it.