AI has finally arrived, and change is coming to life sciences
Developments in AI are set to have profound implications for the life sciences sector, especially in drug discovery. IP professionals will need to adapt their strategies as AI invents, data grows in significance and new types of partnership emerge
The rapid rise of AI means that we will soon be throwing the traditional pharma IP strategy book out the window. New technologies are constantly arising and disrupting industries, including many that have a direct impact on the IP world. IP laws, systems, processes, strategies, offices and, of course, professionals are all susceptible.
Consider the copyright implications of the printing press, TV broadcasting, the personal computer and electronic information storage; the trademark implications of e-commerce, mobile apps and social media; or the patent implications of software, 3D printing and CRISPR. AI will be, and indeed already is, another of these disruptive technologies – and its impact on intellectual property will be profound.
The promise of AI and the many technologies that fall under its broad umbrella have been around for a long time. Recent advances such as increasing computing power and access to data have led to an explosion of practical applications. The rising number of patent applications shown in Figure 1 is one indicator of its growing real-world applicability.
Figure 1. AI patent applications by year of first publication
Indeed, AI can be used to tackle numerous problems. Figure 2 indicates this potential breadth, again using patent filing statistics. Numbers of AI patent applications are increasing across a broad range of fields. It is therefore unsurprising that a wide range of industries have felt the impact of this new technology. Examples of real-world uses of AI include speech recognition in virtual assistants, route planning in satellite navigation systems, natural language translation, smartphone text correction and prediction, chess and Go game playing, autonomous vehicles, energy management, network security threat detection, identity verification and legal document processing solutions. AI is now also beginning to have an effect on life sciences and the pharmaceutical industry.
Figure 2. Patent families for top application field categories by earliest priority year
Given its potential to transform many areas, AI could reshape diagnosis support, treatment recommendations, medical imaging, informatics, screening, clinical trial management, precision medicine, genetics, biological engineering, nutritional and agricultural sciences, and much more.
The future of pharma is tech-first drug discovery
Drug discovery is an immensely challenging problem, as shown in Figure 3. There are more than 9,000 untreated diseases in the world at present and more than 300 million people suffering from rare diseases for which scientists are unlikely to develop treatments any time soon. The drug discovery process still costs an average of $2.6 billion per drug. Even then, between 30% and 50% of the top selling drugs do not work for their patients. Tech-first drug discovery companies tackle this unmet need using AI, developing technology in the service of science, to transform the traditional drug discovery process. On the current trajectory tech-first approaches will become the primary source of novel treatments and success in the field will eclipse traditional methods.
Figure 3. Drug discovery challenges
It is not feasible for an individual scientist to consume the approximately 10,000 scientific papers published every day – and this is only one of the many essential data sources for drug discovery. The AI technologies solving each part of the problem may be wildly different yet most of them work by detecting patterns across huge volumes of data. Scientists can use this technology to make predictions based on patterns in existing knowledge, and AI technology will augment their skills by providing them with new knowledge.
Tech-first approaches can speed up the discovery and development of new treatments by improving each stage of the process. Figure 4 shows how this works for drug discovery. AI can be used to infer what we should know from what is already known. This ensures that researchers start with the right foundations. AI can then provide a more nuanced understanding of the underlying biological mechanisms of a disease, meaning that the causes can be targeted rather than the symptoms. Researchers can also use AI to efficiently generate more effective chemistry, focusing molecular design on developing the right drug. Finally, AI can be used to mechanistically stratify patient groups, increasing effectiveness to well above that 30% to 50% bracket and ensuring that treatments are given to the right patients.
Figure 4. Applying AI technology to the drug discovery process
An end-to-end AI solution with the right foundations, targets, drugs and patients will revolutionise the drug discovery process. This will have a hugely positive effect on society and allow a dramatic increase in speed, efficiency and accuracy in drug discovery – that is, drug discovery with reduced risk and cost. This will be a future where cost-effective medicines are available across a greater range of commercially viable diseases.
Pharma IP strategies will need to adapt
IP strategies will undoubtedly adapt to the emerging mainstream use of AI, as they have for other advances in technology. The AI developments in drug discovery R&D will prompt similarly fundamental changes in pharma IP strategy.
Some of these adjustments will be to compensate for the indirect effects of AI. For example, we will need to change how we protect innovation in the increasingly popular field of precision medicine. This will include managing the changing patentability requirements and the increasing value of trade secrets. We must also consider the potential impact of the geographic stratification of patient populations. There may be scenarios in which a new drug works on 90% of the population in one region and only 10% in another. Here, we may need to refine our strategies for jurisdictional IP rights.
There are three specific changes that directly relate to the use of AI and are of particular interest, namely:
- AI as inventor and the knock-on effect on patent licensing business models;
- an increasing focus on data, particularly IP licensing and ownership; and
- the growing need to collaborate under very different partnership structures.
AI inventions will disrupt business models
AI will invent. In fact, it is already inventing. This was recently demonstrated by the so-called ‘creativity machine’ DABUS. This is not technology thinking for itself, but rather AI being responsible for novel creations that are otherwise patentable today. In other words, it is AI making the kinds of contributions that we consider inventive when made by a natural person.
The law in most, if not all, jurisdictions requires applicants to identify human inventors. Offices or third parties may challenge patent validity in some countries if the inventor list is incorrect – this could include the applicant failing to name an AI inventor or naming a human inventor instead.
The divergence between the emerging AI reality and the law means that patent offices and courts will need to make some novel decisions. Can technology be a named inventor? Is it necessary to consider AI contributions? How should we handle inventions that have no human contributions? Can the ‘user’ (ie, a person or a company as a legal entity) claim inventorship for an AI contribution? Does it make sense to continue to name inventors at all? Patent applicants will also have to make some tough decisions. Should they disclose AI contributions? Should they artificially include humans in R&D processes to ensure a human inventive contribution? What combination of named inventors is appropriate in various scenarios?
Anecdotal evidence from IP industry experts suggests that granted patents for AI inventions already exist. Here, applicants have named human inventors instead of the technology. The Artificial Inventor Project is testing how offices will react to AI inventors. When patent applications for two creations of the DABUS system were filed at the USPTO, UK IP Office (UKIPO) and EPO, the EPO and the UKIPO rejected them. These cases will ultimately be decided in appeal or court proceedings and could set useful new precedents.
There are many arguments for and against allowing AI inventors. There are also many potential implications for either outcome. If an AI creation is identical to something that would be deemed patentable if created by a human, why would one be patentable and the other not?
One of the key purposes of patent systems is to incentivise investment in innovation. Ignoring the long-term possibilities of artificial general intelligence, we are unlikely to incentivise AI technology itself with patent exclusivity. However, protection for AI-enabled inventions would incentivise investment in the development of AI, as feature exclusivity and licensing capabilities are likely to be important to the technology creator. AI technologies have the potential to contribute to society by solving some of the world’s greatest problems. Why would we not want to incentivise that?
Some might argue that patent protection of the AI technology itself should be enough of an incentive. But this protection can be problematic for several reasons. Patentability issues related to software, mathematical methods and abstract ideas all make it difficult to obtain protection. What is more, patent value issues such as infringement detectability reduce the incentives to do so.
We should also consider what might happen if AI inventions were to become patentable. Applicants could flood patent systems as AI invents at scale. Arguably, this level of innovation would prove the success of incentivisation. However, examination resources are already stretched. Prolific single-actor inventors could have competition implications. Should we give AI inventions the same rights as human inventions? Should patent terms be shorter? What changes would we need to put in place? How can we achieve the difficult challenge of international harmonisation? Who should own an AI patent and how do we ensure the appropriate transfer of rights? If we deem AI inventions patentable, what does this do to the definition of a ‘skilled person’? How would this affect inventive step, obviousness, equivalence and sufficiency evaluations? There are many open questions; opinions vary on the answers.
The pharma industry needs patents and the drug discovery business model depends on patent licensing. Due to changes in R&D technologies and processes, drug innovation that is patentable today could become unpatentable in the future – and the consequences of this could be significant.
For AI to become the primary source of new medicines, investment in pharma technology development is essential. If there were no patent incentives for AI-generated treatments, the economics for tech-driven pharma might fail. Such businesses are especially vulnerable in these early stages and need time to generate revenue from technology outputs at scale.
These changes will even affect traditional pharma companies. If AI inventions are not patentable, traditional pharma loses a potential source of new drug products. If the bar for patentability rises, innovators will struggle to protect their own output. With all pharma companies in potential need of alternative protection strategies for their drug products, they might increase their focus on market and data exclusivity, as well as on trade secret licensing. As such, it is possible to envisage a future pharma business model that may not involve patents at all.
Data will become an essential part of IP strategy
Data is an essential component of most AI technologies given that machine learning techniques make predictions based on patterns in training data. The abilities of most AI systems are partly defined by the quantity and quality of the data fed into them.
Although they will remain the primary revenue source for pharma companies, drug products will no longer be their primary assets, but rather will be replaced by proprietary technology and data. In turn, the combination of new technology and data assets will be the key enablers of revenue-earning drug products.
An increasing demand for greater quantities and types of data comes with a wide range of IP implications. Data owners will need to decide if and how to protect their proprietary assets – and if and how to monetise them. Moreover, pharma companies will need to determine how best to acquire data to which they do not currently have access. These decisions will form a key part of pharma IP strategy.
Protecting proprietary data with IP rights can be problematic. While copyright offers some protection, it does not protect facts, so it is unlikely to apply to the large quantities of biomedical data found in structured databases. Database rights also offer some protection for arrangements of data. Yet AI systems are unlikely to store information using the protected database structures in its sources. As such, trade secrets and contractual provisions are likely to be the best armour for proprietary data. But is this enough? Are changes needed in IP systems to provide greater protection mechanisms for data? Would an entirely new IP right be appropriate? These are open questions with which the IP community must grapple.
Sharing data can help others to solve some of the world’s most pressing problems and the mission for pharma companies usually involves improving patients’ lives. Keeping unused data in proprietary silos would seem to go against that. In addition, concerns of missed revenue opportunities encourage inertia – as do the competitive or reputational risks of others extracting value from data that you could not. This is before even considering ethical and privacy issues related to patient data. Many in the industry have attempted data-sharing initiatives, but most of these have been narrow in scope – with limited success.
The available biomedical data is vast and fragmented, as shown in Figure 5. No single company is likely to have access to all or even most of the data that it needs. The data licensing efforts required even within a single area of focus can be significant, and a wide spectrum of data providers and licensing models (shown in Figure 6) complicates this further.
Figure 5. Types of biomedical data used in drug discovery
Figure 6. The data spectrum
This complexity is partly due to the sheer quantity of data sources and the scale of effort required. Other issues that arise include:
- navigating the different types of provider and licensing model;
- ambiguity in licence terms (eg, how data can be used);
- licence compatibility in data aggregation (similar to open source);
- machine processing rights and text and data mining issues;
- foreground rights for different types of processing output (including derivatives, machine learning models and inferences); and
- compliance with licensing obligations (eg, security and internal/product usage restrictions).
Navigating data procurement through this labyrinth of issues requires a niche set of skills. Individuals in-licensing data will need to be a combination of dealmaker, relationship manager, salesperson, requirements gatherer, strategist, commercial and IP lawyer, compliance manager, data subject matter expert and more. Unsurprisingly, this range of skills does not fit into a single obvious role or department. With complex IP issues and the similarities to other types of IP licensing, professionals in the field are likely to be dragged in to help or to lead. Therefore, data licensing will become a key part of both the IP role and IP strategy.
AI will motivate new types of partnership
Pharma companies without sufficient development experience may struggle to develop effective technology in-house. They may also find it difficult to generate value from in-licensed standalone technology.
A tech-first approach to drug discovery requires more than just technology and data. To work efficiently and effectively, it requires a brand new approach to R&D. This includes R&D processes, organisation structure and internal culture. It requires a close integration of technologists and scientists. Successful pharma companies will be those that appreciate the sizeable effort needed to transform their businesses. Underestimating this may lead to expensive and time consuming development and procurement failures. Those that do not keep up in the technology race risk falling back on other strengths as well. Failure could mean focusing on large-scale commodity clinical trial management and drug distribution.
To gain early access to the low-hanging fruit of tech-first drug discovery, pharma companies will need to collaborate. The list of partnerships beween pharma and technology companies in Table 1 shows that this is already happening. In fact, this is only a sub-set of the many collaborations announced in 2019.
Table 1. Selection of 2019 pharma AI drug discovery collaborations
However, the close integration of cutting-edge technology and hugely complex science causes new combinations of contractual issues to arise:
- IP rights: Who gets the rights to the drug’s intellectual property? What about the intellectual property that the collaborators decide not to progress? AI automation may generate a significant amount of new drug-related intellectual property. Should the tech company be able to keep anything not taken forward for development? What about improvements to technology, as well as data and other arising intellectual property? Should any of these IP rights be exclusive? Should we limit any of them to a field of use? Pharma companies will not want their competitors entering similar markets, but tech companies will need to collaborate with many pharma companies. How can we define exclusivity fairly given the conflicting requirements?
- Financials: What levels and combinations of up-front fees, milestones, royalties and FTE costs are appropriate? What is fair compensation given the contributions of both parties, the future drug development plans and the risks and potential upsides?
- Security: Who controls and has access to the infrastructure hosting the technology? Who pays the costs? What information security measures are in place? Who can use the technology? Are audit rights needed, before or after the collaboration? Does this provide adequate protection for the tech company’s proprietary technology, or for the pharma company’s proprietary data?
- Data licensing: A large quantity of data will likely come from third parties beyond the collaborators. Who will obtain this data? Can they get appropriate rights to share the data so that the collaborator can process it? Will data acquisition timelines fit with the collaboration plan?
- Complexity: These collaborations are particularly complex. It will take time for each side to get up to speed with the requirements and capabilities of the other. This is especially problematic if parties bring in individuals to draft contracts towards the end of the process.
These new technology and pharma collaborations will be very different from traditional industry partnerships. While tech companies will be extremely protective of their technology, pharma companies will be highly protective of their data. What is more, both will want a fair share of the downstream monetisation.
AI is changing the world, including drug discovery and intellectual property. The pharma IP strategy of tomorrow will look very different from that of today. To continue to be effective, IP professionals will need to adapt. It is an exciting time to be working in pharma.
The impact of AI on pharmaceutical IP strategy means that IP professionals will need to adapt. Three specific actions are recommended for IP leaders in pharma companies:
- Train staff on AI technologies, development practices and business models.
- Plan for a world without patents.
- Recruit employees with tech experience into senior positions.
Training budgets should also be spent on data licensing. This could include:
- the relevant types of data, types of provider and usage rights needed by the business;
- how copyright and database rights apply to data;
- the relationship between AI and data, including the flow from raw data to derivatives, machine learning models and output predictions;
- how open source communities have tackled issues such as licence compatibility and management; and
- policymaking activities such as open access, text and data mining, and data licence standardisation.