Artificial Intelligence (AI) has created music, works of visual and literal art, and new languages. And now it is creating new pharmaceutical drugs.
In 2010, a report by the Tufts Center for the Study of Drug Development estimated costs to develop an approved drug to be $2.6 billion and requiring an average of 10 years to come to market. Now, a UK based company whose entire valuation is less than that figure, and which has only been existence for 5 years, has 24 drug candidates in its pipeline. The secret to this accelerated development? AI.
AI is changing the world as we know it. A recent report by PWC predicts that global GDP will increase 14% or nearly $16 trillion dollars by 2030. The report concludes that the healthcare sector is likely (along with the automotive industry) to be impacted the most by AI.
One company, Benevolent AI, is beginning to show proof of the impact that AI can have in the pharmaceutical industry in particular. It uses AI to scour the wealth of scientific information to identify potential drug candidates, develop new molecules, and create novel hypotheses for therapeutic targets. And it is doing it at lightning speed.
Since its creation in 2013, Benevolent AI has developed a portfolio of 24 potential drug candidates. Some of these it licensed to other companies for development, particularly when it first came onto the scene. But Benevolent AI did not want to be pigeon-holed to a drug-discovery role. It wanted to push drug development from start to finish.
This led to its recent acquisition of a drug discovery and development facility on the Babraham Research Campus in Cambridge (UK) adding capabilities such as assay development and screening, medicinal and synthetic chemistry, drug metabolism and pharmacokinetics, pharmacology and clinical development. This will allow the AI company to control the process of drug development from discovery to clinical development, to manufacturing.
Now, just 5 years after its inception, Benevolent AI has one drug in phase IIb clinical trials looking at efficacy and safety for treating Parkinson’s disease. This drug was one compound from a series that was licensed from Janssen Pharmaceutica NV, a member of The Janssen Pharmaceutical Companies of Johnson & Johnson.
Also generating excitement is a compound that Benevolent AI has identified for the treatment of amyotrophic lateral sclerosis (ALS) or “Lou Gehrig’s disease”. If approved, this drug could be one of the first drugs whose entire development is touched by AI in some fashion. The compound was 1 of 5 novel compounds that the company’s AI identified as having potential for treating ALS, a complex neurodegenerative disorder that leads to death of motor neurons. Preclinical studies conducted in collaboration with the Sheffield Institute of Translational Neuroscience found that the compound could prevent neuronal death and slow disease onset.
Benevolent AI’s success in leveraging AI technology is encouraging and with AI taking centre stage more broadly, it is inevitable that other companies are bound to follow and drive innovation in the industry.
It is easy to see why AI can make such a large impact on the drug development process. According to Benevolent AI, 90% of the world’s data has been created in the last 2 years. Scientific publisher Elsevier published approximately 400,000 new peer-reviewed articles in 2015 alone. This represents only 16% of the total number of scholarly articles worldwide in 2015. A study from the Max Planck Society in Munich, Germany and the Swiss Federal Institute of Technology in Zurich estimates that scientific information (not just publications, but also books, websites, and other outputs) is growing at an exponential rate, with the total output of scientific information doubling approximately every 9 years.
No human could process the amount of information created in one year, in their entire lifetime. Enter AI. As an example of how companies are leveraging the recent rise in machine learning and AI developments, Benevolent AI was able to train its AI to learn from every chemical reaction ever published (approximately 12.4 million). The AI is now using that training to predict and plan synthesis of new molecules. And that is only one source of information that Benevolent AI’s tech is drawing from.
But with all of the money pouring in and the potential for AI to revolutionize the industry, one of the key considerations for investors and companies alike will be how to monetize these innovations. What types of IP protection are available for AI influenced drug discoveries to offset costs of R&D? With average drug development costs potentially as high as $2.6 billion, how can companies generate profits which can be reinvested into continuing drug development.
Pharmaceuticals have typically lent themselves to patent protection. Given the nature of the industry, that is unlikely to change - drugs do not fit within the scope of copyright protection. Given their ability to be reverse engineered, and the degree of disclosure and transparency required for regulatory approval, trade secret protection is often insufficient or impractical. But is patent law ready for AI developed drugs?
One concern about patent protection is the issue of inventorship. The Canadian Patent Act does not define “inventor”, so on its face, if a machine had conceived of and developed an invention in its entirety, that may not preclude patent protection. However, some references to “inventor” in the Patent Act imply acts that only humans could perform (e.g., in respect of assignments, who can assign a patent right, when, and how, suggest that only humans are contemplated as assignors). In the US, the Copyright Office has taken the position that to be registered, a copyrighted work has to have been “created by a human being.”
The Copyright Office’s policy has been affirmed in courts as well. Just this week, the 9th US Circuit Court of Appeals affirmed a lower court ruling that animals (i.e., non-humans) do not have standing under the Copyright Act and cannot own copyright, and dismissed a lawsuit over the famous Naruto monkey selfie. Similar issues will inevitably arise in patent disputes and patent offices in both the US and Canada may soon find a need to provide clarification on requirements for “inventorship” similar to what has been done in the copyright context. Courts will no doubt have to grapple with these issues as well.
In the absence of clear guidance, prudent practice would suggest that including a human in the invention process (e.g., by selecting a candidate from several potential targets generated by AI) could alleviate this potential issue as they would be the “inventor” for purposes of the Patent Act. This is sometimes referred to as “keeping a human in the loop”.
By partnering with Sheffield, Benevolent AI is doing just that with its ALS drug. Indeed, preclinical testing revealed 4 of the 5 candidates identified by the AI to be ineffective or no more effective than standard of care. The vetting of the 5 molecules identified by the AI required human influence in preclinical testing. Thus it appears that, at least for now, human involvement is still required at some step in the development of a drug.
One key consideration with “keeping a human in the loop” is to ensure that the human is acting in a way that elevates their course of conduct above what would be considered that of a person of ordinary skill in the art (POSITA) in an obviousness analysis.
Obviousness generally will become a hotter point of contention moving forward too. Key issues will include whether an AI machine can form part of the definition of POSITA? AI that only automates tasks might be considered part of the POSITA, but it could be argued that AI that learns and creates new hypotheses possess the “ingenuity” or “inventive spark” characteristic of traditional inventors.
The “obvious to try” test will likely require some refinement as well. As AI becomes “smarter” and more powerful, how will concepts such as “undue burden” transform? If AI is part of the definition of a POSITA, if something is “obvious to try” to a human, but not to advanced AI, would that meet the burden? Conversely, if AI can come to the invention in weeks where it would otherwise take a human years to come to the same invention, is the invention obvious?
Utility is another issue that is sure to be impacted by AI. In Canada, an invention’s utility must have been either demonstrated or “soundly predicted” as of the Canadian filing date. As AI continues to learn through experience and exponentially expanding data sets, it might be argued that results from AI tests could substitute for in vivo testing. This would likely not be sufficient for the purposes of regulatory approval, but in a patent dispute, it could be argued that in silico testing (i.e., simulation by a computer) provides at least a sound prediction of utility, if not outright demonstrated utility. Reliance on in silico testing to establish utility could also affect obviousness – if AI could soundly predict the utility of an invention, would that make the invention obvious to a POSITA (particularly if AI can form part of the definition of a POSITA)?
Additionally, it will be important to ensure that evidence of utility can be extracted from AI and that at least this portion can be “white-boxed” so that the way in which the compounds were tested and the results themselves are extractable.
These questions may at first seem academic, but companies like Benevolent AI are poised to transform the industry and are already discovering new drugs with the assistance of AI. Present leaders in “big pharma” are developing in-house AI departments and investing and exploring AI technologies. It may not be so far-fetched to think that courts, legislatures, and intellectual property offices will have to address these developments in the pharmaceutical industry in the near rather than distant future.
It will be imperative for companies to stay ahead of these issues and anticipate litigation related risks well in advance by:
- Considering IP strategy early and managing IP portfolios accordingly (e.g., timing of filing in light of potential impacts on utility and obviousness);
- Keeping apprised of regulatory positions on AI and IP rights (e.g., staying informed about any guidance from the United States Patent and Trademark Office or the Canadian Intellectual Property Office on AI-influenced inventions); and
- Modernizing commercial agreements to help protect IP and proactively address potential issues that may arise once AI becomes the substance of litigation (e.g., inventorship rights and assignments, ability to “white-box” evidence of utility).