AI enabled drug discovery and design

This article is part of our Biotech Review of the Year - Issue 12 publication.

2024 has seen another wave of headline grabbing AI drug discovery deals with multi-million (and even billion) dollar biobucks. Here we look back at some of the recent trends we have seen in AI for drug discovery and design and highlight some key points to consider when drafting and negotiating AI enabled collaborations. 

Significant developments in computing, and the rapid development and improvement in AI capabilities in recent years means that we are now at a stage where pharma and biotech companies are routinely making use of AI in a broad range of applications (including from clinical trial design and recruitment of clinical trial participants to generating regulatory submissions and compliant marketing materials). However, it is the potential applications of AI in drug discovery and design which have been generating the most excitement in the field. 

The costs of bringing a new drug to market are astronomical, with recent estimates placing costs up to $2.8bn. These costs, coupled with development timelines in the range of 12-15 years from initiation of a discovery programme to approval, highlight the opportunity and need for innovative technologies to transform the drug discovery process. As if that wasn’t compelling enough, in addition to promised time and costs savings, AI also has the potential to support target identification and precision drug design leading to better optimised drugs targeted to novel targets.

Healthy deal activity 

This combined promise of better, quicker and cheaper drugs has been driving AI for drug discovery dealmaking in recent years. While at the time of writing, the total number of partnering deals and deal values for companies engaged in AI enabled drug discovery and development during 2024 do not look likely to meet the market highs of 2021, there has still been a steady stream of partnerships between major biopharma companies and AI-focused companies. Examples of 2024 collaborations include Novartis’ deal with Generate:Biomedicines to use the latter’s generative AI platform to discover and develop protein therapeutics (overall deal value of more than $1bn); Eli Lilly’s collaboration with Genetic Leap to use Genetic Leap’s RNA focused AI platform to generate oligonucleotide drugs against specific targets (overall deal value of more than $409m); and Nxera Pharma’s collaboration with Antiverse to use Antiverse’s generative AI platform to design antibodies for G-protein coupled receptors (GPCRs) – targets which have traditionally been very challenging (undisclosed deal value). 

However, recent research carried out by PwC indicates that while AI drug discovery partnerships often include eye catching overall financial sums, in many cases financial allocation is shifted towards milestone payments, with deals including lower upfront payments. Pushing value realisation into milestone and royalty payments which only become due once a product has demonstrated a measure of success is a typical way of de-risking a deal and indicates that biopharma companies are wary of taking on too much risk with any one company or deal. Instead, many biopharma companies are striking multiple drug discovery partnerships with different AI-focusedpartners. Given the still relatively nascent state of the AI field and the proliferation of AI-focused companies with different types of platforms, capabilities, and areas of focus, this is perhaps unsurprising. 

Companies who are pursuing a multi-collaboration approach with different partners will need to consider carefully how those partnerships are managed, what indications or targets they are aimed at (and whether there is any risk of overlap, particularly if any of the partnerships involve a degree of exclusivity) and whether and how proprietary data is being shared with those partners. 

As well as collaborations, 2024 also saw consolidation in the sector (eg the merger between Utah headquartered Recursion and UK based Exscientia) and M&A (eg the acquisition by Ginkgo Bioworks of Patch Biosciences which has an AI platform for sequence design). We also saw major biopharma companies continue to make strategic investments in AI companies, such as the September 2024 investment by Novartis in Generate:Biomedicines which was announced at the same time as the target collaboration mentioned above to discover and develop protein therapeutics with generative AI. 

For pharma and biotech companies who are collaborating with AI providers, M&A in the sector can represent a risk, particularly if an AI provider is acquired by a pharma competitor. Pharma companies who are alive to this risk may seek to address change of control specifically in their partnership agreements. Key points to consider include (i) early termination rights and (ii) data segregation or destruction obligations applicable to the pharma company’s proprietary data, results of the collaboration and other confidential information. Depending on the nature of the collaboration and the AI model used, some pharma companies may also seek to guard against a competitor gaining access to models which have been trained on and improved by use of the pharma company’s proprietary data by requiring the AI provider to create a separate instance of its platform to be trained on the pharma partner’s data and can, if required, be segregated or destroyed in the future. 

Generative AI leading the way

Readers will be aware of the explosion in the use of generative AI tools which has taken place over the last two years since the 2022 launch of the large language model ChatGPT. In last year’s Biotech Review of the Year we examined five of the key issues for life sciences organisations considering incorporating or leveraging generative AI. We would encourage readers to revisit that article which remains just as applicable today (and addresses issues such as confidentiality of inputs, patient privacy and security, IP ownership and infringement risk and ensuring accuracy and quality of outputs). 

Generative AI can be used not only for generating content such as text and images but is also increasingly being applied in drug discovery. Particularly exciting is the potential use of generative AI to design antibodies for challenging targets (such as G protein-coupled receptors) or to design best in class drugs for existing targets. The large amount of publicly available protein structure data, combined with the recent innovation in AI model capabilities has had a revolutionary impact on the ability of AI to model the structure of complexes including proteins, nucleic acids, small molecules and ions (such as DeepMind’s neural network based generative model AlphaFold 3). A significant number of companies are now pursuing a generative approach to antibody design.

As discussed previously, where a generative AI tool is being used to generate novel molecular structures (such as antibodies) and the model is being trained on an organisation’s confidential data (eg existing structures and target data), there is a risk the tool could be used to reproduce the same structures for competitors. As mentioned above, some companies are taking the approach of requiring AI partners to host a separate instance of their tools which can then be segregated or destroyed. However, this can be hard to negotiate and our experience is that it is common for AI providers to seek rights to continue to use data generated in the collaboration for the continued training and improvement of their platforms. It does not follow however that the AI provider should have the right to retain and use a company’s proprietary background data beyond the collaboration. Before embarking on a collaborative partnership with an AI provider, companies should consider technical due diligence and work with R&D teams to establish what data will be exposed to the tool and consider whether and how that data should be protected.

Modalities: biologics 

While the use of AI-driven systems that focus on design and optimisation of small molecule drugs (whose characteristics and behaviours are better understood than biologics) is relatively widespread at this point (including AI-driven screening tools), the use of AI for the design and optimisation of biologic drugs such as antibodies has been accelerating more recently, with much of this activity concentrated in AI-focused biotech and biopharma companies. 

Notably a number of 2024’s major AI collaborations have been focused on biologic drugs (such as the Novartis Generate:Biomedicines partnership mentioned above and Eli Lilly’s partnerships with Genetic Leap and Open AI). Each of these collaborations also leverages generative AI for antibody design, indicating a deal trend we expect to continue.

Access to data and changing business models

AI models must be trained on large amounts of reliable and structured data. With limited public data available in some areas and the vast stores of data held by big pharma not readily available (or necessarily structured in a way which is optimised for training an AI model), a growing number of AI-focused biotech are setting up their own wet labs to generate data to train their AI models. This represents a business model pivot for many companies which were previously computer based only and backs up the trend which we have seen in recent years with many AI-first companies looking increasingly like traditional biotech companies, some with their own pipeline of drugs. In particular, we are seeing a growing number of AI companies who are collaborating to design and optimise antibodies using generative AI setting up their own wet labs and developing their own internal pipeline of antibodies in development (such as Absci and BigHat Biosciences). This diversification strategy provides not only another source of data (from the AI company’s own wet lab) but also another potential source of revenue if pipeline products are successful. It does mean however that pharma and biotech partners need to be mindful that the AI company could be a potential competitor itself. 

Careful thought should be given to the field of the collaboration and pharma companies may seek exclusivity and (time limited) non-compete restrictions in their collaboration agreements with these companies to protect the competitive advantage they hope to gain by using the AI platform in the first place. Legal advice should of course be sought as to the enforceability and competition law risks regarding non-compete provisions. 

Summary

The use of AI in drug discovery is not new, but recent rapid advances in technology (such as generative AI) and evolving business models of AI-focused companies has led to a continued trend of AI-driven R&D collaborations. We are seeing biologics (particularly antibodies) increasingly become the focus of these collaborations. While many of the terms of these collaborations will feel familiar to those well versed in more traditional drug discovery collaborations, the evolving AI market (including the likelihood of M&A) and the ways in which these powerful AI platforms use data, means there is also plenty new to consider when drafting and negotiating AI-enabled collaborations.