How to use computational chemistry results in patent applications

To what extent can inventors and companies use computational chemistry results to strengthen their patent applications?

In the chemicals industry, both theory and experiments are often used in research. Oftentimes, both types of results make it into the patent applications.

While computational chemistry can be helpful at all stages of the patenting process, it’s not as routinely used as larger-scale methods such as CFD and process modeling.

Specifically at the drafting stage, here’s two ways computational chemistry can be used toward expanding claim scope:

First, for a given invention, such as a particular molecule or material, calculating variations “around” that invention can show which ones are likely to have similar performance, helping describe and enable additional subject matter. Usually, many, many more systems can be characterized computationally than experimentally.

Second, computations can identify the underlying, abstracted properties behind an invention’s superior performance, which can then themselves be claimed.

This is the famous “quality common” of patent enablement, first described in Thomas Edison’s incandescent lightbulb case of 1895. There, the quality common was a specific parallel arrangement of woody fibers in the lightbulb filament—this discovery allowed the bulb to glow for a long time without charring, ushering in the electric era.

Recently, the “quality common” was also discussed by Justice Gorsuch, in the unanimous 2023 SCOTUS decision on patent enablement. In Amgen v. Sanofi, Amgen tried to claim a universe of antibodies based solely on their functional properties, without identifying any amino acid sequences that cause that function. While these claims were invalidated, Gorsuch noted that, even without sequence information, Amgen may still have been entitled to very broad claims if it had identified some “quality common” underlying every single antibody it was trying to patent. Unfortunately, he didn’t provide any examples of what such a quality common could be in antibodies.

In chemistry inventions, what could such a “quality common” look like, and how can it be used?

Here’s a completely hypothetical example. Say the discovery is a small molecule that’s recognized by some particular receptor (an enzyme pocket, or a molecular sensor). Naturally, the inventor also wants to patent structurally similar molecules that should also interact with that receptor. The molecule has separated electron-rich and -poor functional groups, which form a dipole that happens to perfectly complement a dipole on the receptor. The presence of this dipole on the molecule is the underlying “quality common” that achieves the molecular recognition—not necessarily the specific atoms and functional groups themselves. In a patent application, describing and claiming this dipole effect can be another way of enabling broad claims, compared to the more traditional approach of just listing a number of electron-donating and electron-withdrawing groups, and trying to claim all combinations.

In my experience, these types of underlying mechanistic reasons are pretty much always thought about in R&D, the difference being the extent to which they’re ever validated with calculations or additional experiments.

What about the “accuracy” of the computational chemistry?

A related question is how computation-enabled claims might stand up to scrutiny at the Patent Office and later in litigation. These questions would likely be highly fact-intensive, with the hypothetical “Person of Ordinary Skill in the Art” (POSITA) playing a central role.

Applicants should use sufficiently high-quality and widely-accepted models. Particularly, models shown in the scientific literature to be accurate for similar systems can be especially useful for indicating that a POSITA would find reasonable correlation to experiments. Applicants can also conduct their own benchmark studies against simple systems to validate their models. Finally, applicants should use a mix of theoretical and experimental data to form a convergent story that supports broad claim scope.

The above discussion has in mind physics-based models such as DFT and classical MD and Monte Carlo. In the coming years, new generative AI models for chemistry and materials science may make all these questions even more relevant.