The pharma and biotech industries have embraced the use of artificial intelligence (AI) in developing new drug candidates and new uses for known drugs, devoting resources to the use of AI in technology development.

For example, in August 2017, it was announced that GlaxoSmithKline (GSK) had entered into a collaborative relationship with Insilico Medicine, a Baltimore, Maryland-based AI company. Insilico focuses on applying AI to drug discovery, biomarker development and aging research. Similarly, Pfizer has incorporated the use of AI as a part of its R&D efforts in its Genome Science and Technologies group.

The question generally foremost in people’s minds concerning the use of AI to develop and invention is whether an AI can be an inventor.

Another important consideration is whether an invention directed to a new pharmaceutical compound or the new use for known compound, which has been developed using AI would be considered prima facie obvious based on the use of the AI?

Case study

In the scenario under consideration, AI is used to screen databases of potentially thousands of compounds to identify a new compound or a new use for a known compound. Thus, the invention the invention has arguably been made by running a computer algorithm to screen known data.

In 2009, the Court of Appeals for the Federal Circuit decided in In re Kubin (2009). In the Kubin case, the claims were directed to nucleic acid sequences encoding natural killer cell activation-inducing ligand (NAIL). The claims were rejected as being obvious over the combined teachings of US patent no. 5,688,690 “Valiante” and the general cloning manual, “Molecular Cloning: A Laboratory Manual” (Sambrook et al). Valiante disclosed a receptor protein called “p38”, which is the same protein as the NAIL protein.

The question in Kubin was whether any gene sequence encoding NAIL would be prima facie obvious once the protein itself is known. The court applied the analysis set forth in KSR International v Teleflex (2007) and stated that: “When there is a design need or market pressure to solve a problem and there are a finite number of identified, predictable solutions, a person of ordinary skill has good reason to pursue the known options within his or her technical grasp.

“If this leads to the anticipated success, it is likely the product not of innovation but of ordinary skill and common sense. In that instance the fact that a combination was obvious to try might show that it was obvious under §103.”

The court then considered, when is an invention that was “obvious to try” nevertheless non-obvious? The court stated that there are two instances where “obvious to try” does not equate with “obvious”.

In the first instance while it might be “obvious to try” by varying all parameters or trying each of numerous possible choices until one possibly arrived at a successful result, “where the prior art gave either no indication of which parameters were critical or no direction as to which of many possible choices is likely to be successful” the invention would not be obvious.

The court stated that: “where a defendant merely throws metaphorical darts at a board filled with combinatorial prior art possibilities, courts should not succumb to hindsight claims of obviousness.”

However, the court also noted, again citing to the KSR decision that: “where a skilled artisan merely pursues known options from a finite number of identified, predictable solutions, obviousness under §103 arises.”

The second class of an impermissible “obvious to try” situation occurs when “what was ‘obvious to try’ was to explore a new technology or general approach that seemed to be a promising field of experimentation, where the prior art gave only general guidance as to the particular form of the claimed invention or how to achieve it”.

The court balanced this situation against KSR and further noted that, on the other hand, an invention would be obvious if the improvement achieved by the invention were no more than the predictable use of prior art elements according to their established functions.

USPTO’s position

Subsequent to the Kubin decision, the author of this article was advised by the Patent Trial and Appeal Board (PTAB) during an appeals hearing that the US Patent and Trademark Office (USPTO) position is that once a genome has been sequenced, every possible sequence (eg, probe, primer, EST, etc) is per se prima facie obvious because every sequence can be routinely and predictably derived using a computer algorithm, regardless of how many millions of sequences that may be.

It is conceivable that a similar analysis to Kubin may be used in determining whether a new pharmaceutical invention developed using AI would be considered prima facie obvious because of the AI used in its development. The questions are whether the use of AI in developing a new pharmaceutical compound would be considered conventional and whether the results obtained would be considered predictable?

The more difficult question to answer may be whether the new compound would be considered “predictable” based on the analysis used in Kubin and based on the current position of the USPTO with regard to sequences obtained from a known genomic sequence.

The court in Kubin noted that “obvious to try” does not equate to obvious situations where it is required to “vary all parameters or try each of numerous possible choices until one possibly arrived at a successful result, where the prior art gave either no indication of which parameters were critical or no direction as to which of many possible choices is likely to be successful”.

The court analogised this to throwing darts at a dartboard hoping to hit something of use. When an AI used to screen databases of thousands of compounds to design new compounds, it may be said that the AI is varying all parameters and trying all choices until a compound of use is achieved. Thus, the new compound is not obvious.

However, a counter argument may be made that the prior art does provide guidance in the form of common structural elements in compounds, known activity data, etc, and it is this guidance which allows the AI to make design choices based on the databases of compounds.

In addition, as noted above, the position of the PTAB is that once a genome is known all the possible derivable sequences are prima facie obvious because all possible sequences may be identified using a computer algorithm. This same analysis may be applied to using AI (a computer algorithm) to identify new compounds from databases of known compounds (ie, the equivalent to a known genome) based on the logic that: “where a skilled artisan merely pursues ‘known options’ from a ‘finite number of identified, predictable solutions,’ obviousness under §103 arises”.

However, the court also noted that an impermissible “obvious to try” situation arises where “what was ‘obvious to try’ was to explore a new technology or general approach that seemed to be a promising field of experimentation, where the prior art gave only general guidance as to the particular form of the claimed invention or how to achieve it”.

One of the key elements in finding the sequences obvious in Kubin was that the techniques used to obtain the sequences were well-known and conventional. AI on the other hand, is not, at this time, well-known or conventional. AI is a new technology as applied to drug design and at best the prior art only gives general guidance as to how to achieve the new invention.

Thus, currently, the development of a new pharmaceutical compound or new use of a known compound should not be considered prima facie obvious simply because AI was used to develop the new compound or new use.

Summary

As AI technology progresses there are several considerations that may come into play in finding whether a new pharmaceutical compound or new use of a known compound is prima facie obvious because AI was used to develop the new compound or new use.

One important consideration may be whether the AI is publicly available. If the AI used is publicly available and commonly used it may become the equivalent to the Sambrook molecular cloning manual. Thus, an important question will be whether the AI is a conventional tool, or is the AI itself new and not in the public domain?

Similarly, a relevant question may be whether the screened data is in the public domain. It cannot be considered the pursuit of “known options” if the database screened by the AI is not publicly available.

As biotechnology techniques have advanced and become more predictable in their outcomes, it has become more difficult to establish the non-obviousness of new inventions in biotechnology based on analysis as discussed in the Kubin case.

The use of AI in the development of pharmaceutical inventions is still in its infancy. However, as the use of AI becomes more commonplace (ie, “conventional”), the bar for showing the non-obviousness of a pharma invention made using AI may similarly become higher.

MaryAnne Armstrong is a partner at Birch Stewart Kolasch Birch. She can be contacted at: maa@bskb.com

Images, from top: Shutterstock / Besjunior, ShotPrime Studio, Carsten Reisinger, gopixa, charles taylor