Article
Why AI drug discovery deals are different from traditional pharma collaborations
The cost and time of bringing a new drug to market continues to rise.
In response, pharma companies are increasingly turning to AI in a bid to streamline aspects of the drug discovery process, from identifying druggable targets and designing candidate compounds, to designing clinical trials and managing supply chains.
With several AI-designed compounds now in clinical trials, pharma companies no longer regard AI as a throw of the dice.
AI’s ability quickly to analyse vast volumes of data and identify correlations is dramatically increasing the efficiency – and reducing the cost – of drug development.
This trend is driving transactional activity across the spectrum of “buy/build/collaborate”, with pharma companies entering into partnerships with AI companies as well as seeking to acquire AI businesses and build out their own team of data scientists.
In recent years, investors have backed hundreds of AI startups in the hope their systems could help discover the next blockbuster drug.
What is ‘artificial intelligence’?
At a high level, “artificial intelligence” is most typically associated with machine-learning, which is itself an umbrella term that applies to different technologies that revolve around computational approaches seeking to emulate (parts of) human decision-making.
A machine learning model is a mathematical model trained by inputted data to predict a solution to a specific problem. The more data inputted, the more accurate its predictions.
In the pharma space, AI has a multitude of potential applications, including identifying which biological targets – such as proteins and nucleic acids – are prevalent in particular diseases, and which chemical compounds could bind to them.
These compounds change the target’s behavior (some more effectively than others) and are used as the basis for new prophylactic or therapeutic treatments.
Collaboration arrangements grow more sophisticated
AI has been playing an increasingly important role in drug development for the past decade, and today the pipeline of AI-designed drugs is growing at 40% a year.
Collaborations between pharma companies and AI companies can take various forms. At one end of the spectrum (and most common in the earliest collaborations), they were often structured as service agreements.
The AI company would suggest a series of candidate compounds, hand them over to the pharma company for development and walk away. From the pharma company’s perspective, the AI was contributing to its process rather than its product.
Today the market – just like AI itself – has changed (and is still evolving), with a more complex array of relationships between parties.
Collaborations are typically longer-term arrangements structured around a “design-make-test” cycle, with the AI company recommending candidate compounds, the pharma company testing them in the “wet lab”, and the results of this research feeding back into the AI model.
AI companies’ role in product lifecycle expands
Taking this further, some AI companies are now involved in more of the product lifecycle have wet lab facilities and product pipelines of their own, and licensing their compound-related IP to pharma companies.
Others have built up their regulatory and distribution expertise to the point where they are in-licensing the rights to commercialise drugs themselves.
Pharma companies, too, are converging on the tech space, recruiting “digital chemists” and in-licensing AI platforms on which to run their own models.
What determines commercial terms?
The degree to which the AI company remains involved in the ongoing development of the product is one of several key factors that will influence the commercial terms of the collaboration.
Others include the objective of the collaboration (ie what the AI is being used for), the initial contributions being made by the parties, and whether the AI company is building a new platform or deploying an existing platform.
These factors will coalesce into key commercial questions, including: (a) should the AI company receive a long-term share of the future upside derived from the product (eg via milestone payments and royalties); (b) is the AI company subject to any exclusivity commitments; and (c) who should own the IP rights in the data generated under the collaboration?
How to manage the culture clash between pharma and tech
While pharma and AI have a lot to gain from working together, the two industries are not natural partners.
Biotech companies move at speed and are typically more agile, while pharma businesses are heavily regulated and tend to have a raft of internal processes designed to ensure quality, safety and efficacy of drugs. Tech companies are not used to operating in such a minefield of regulation and bureaucracy.
Managing this cultural dissonance is critical to the success of any collaboration.
Say the AI company is paid on the number of candidate compounds taken into the lab for testing. What if it generates recommendations more quickly than the pharma company’s existing processes can handle?
What if the pharma company’s internal controls prevent it from sharing data quickly and securely with third parties?
There are also more practical issues to consider, such as whether the pharma company’s decision-makers need guidance on how the AI system arrives at its predictions.
These governance issues must be addressed in the contracts if any collaboration is to succeed.
Data as an asset
During contract negotiations, IP and data are often the focus of intense discussions, particularly in relation to the ownership and usage rights of the data generated by the AI company.
Crudely speaking, in recommending a candidate compound, the AI company will generate two items of potential value – the chemical structure of that compound and the correlations in the underlying data that were used to recommend it.
Collaboration agreements will typically assign ownership of the compound to the pharma company, but what about the rights in this newly structured data?
This has commercial value for both parties, and presents a competitive disadvantage to the pharma company if the AI company uses it to help a competitor in the same field.
But, if a pharma company is benefiting from the learnings of an AI model that has evolved through partnerships with other partners, surely it should too be sharing its data?
The solution to this issue is typically found in a more granular and nuanced approach to different types of data, and in focusing less on the question of ownership and more on the appropriate parameters within which each party is permitted to use the data.
Careful structuring required to protect AI system
Then there is the AI platform itself, which, at a high level, is made up of three components – the model, the data used to train and test it, and the software on which it is programmed.
In most jurisdictions, the AI company will need to rely on various forms of IP right to protect different parts of the system.
In (perhaps limited) cases the AI model might be protectable as a patent, elements of the software (but not the model itself) might be protected by copyright, and the database might be protected by database rights.
The IP rights most likely to protect AI platforms (and each of the components comprised in them) are trade secrets, which protect certain confidential information.
Trade secrets – and how to protect them
In many jurisdictions, the test of whether a trade secret applies is evidentiary, as to whether the necessary steps have been taken to ensure that the information has been kept secret.
In an AI/pharma collaboration, the AI company will want to ensure that the contract only requires it to make available the outputs of the AI model and not the AI model itself.
In a traditional pharma collaboration, each partner typically grants the other a license to any existing (or background) IP that is “necessary or reasonably useful for the collaboration”.
What the AI company doesn’t want is for its platform technology (the cornerstone of its commercial proposition), and any improvements to it, to fall within any such license.
Financial terms evolve as market matures
As the market has matured, the financial terms of collaboration deals have evolved. Early tie-ups often involved fixed financial structures, with the AI company receiving set fee at regular intervals.
More recently, the AI company is typically rewarded in the development phase for hitting development and regulatory milestones (eg MHRA, EMA and FDA approvals) and, once a product is on the market, milestones and royalty payments based on net sales of the products.
It is also increasingly common for partners to share the risks and returns. The relationship might involve cost-sharing during the development phase, while post-commercialisation, the AI developer might get larger milestone/royalty payments – or even a cut of total revenues.
Focus on AI’s contribution to final product
The drug development process can take years, and by the end it can sometimes be hard to link the AI company’s contribution to the final product.
Years ago, when these deals were in their infancy, the contract terms might have allowed the AI company to claim royalties for any successful discovery during a set research term, regardless of whether the product was based on one of the AI’s recommendations.
Now, terms often require a special test to determine how closely the AI-recommended compound relates to the drug that eventually hits the market.
Pharma companies have been working with AI (in various guises) for a decade or more, and they are still scratching the surface of what it can achieve. Structuring their collaboration deals with care will help them realise its full potential.
This content was originally published by Allen & Overy before the A&O Shearman merger
Related capabilities