AI – commercial models and value recognition in life sciences transactions


First published in our Biotech Review of the year – issue 9.

The use of AI will become widespread in the sector

The use of AI and machine learning (ML) technologies in life sciences is already here, but it’s set to become widespread within the next decade. The life sciences industry has attracted some criticism for lagging behind other sectors when it comes to adopting these technologies – although some say this is for good reason, given the consequences in healthcare of “getting things wrong”. Today these technologies are rapidly gaining traction in the sector, with most companies now having AI projects in progress.

AI systems are being applied to all stages of a drug’s lifecycle, from drug discovery and clinical trials, through to post-approval activities such as patient monitoring and market analysis to help optimise sales. However, perhaps unsurprisingly, it is in the area of drug discovery (including identifying suitable biological targets for drugs to act on) where AI tools have been the most widely available and adopted so far. This is partly due to the lower regulatory hurdles in drug discovery (compared to those in clinical development), and the more advanced nature of some of the AI platforms that currently are being commercialised in this space. This is particularly so for AI systems that focus on small molecule drugs, whose characteristics and behaviours are better understood than biologics for example, and for which large amounts of quality data, collected over many years, is available to draw upon to train the AI.

Opportunities to bring down the inflating costs of drug development

These new technologies present a real opportunity with which to speed up the drug development process and also to tackle the inflating costs. This is welcome news against a backdrop of the increasing pressure being put on the industry to make healthcare more affordable. A prescription drug typically takes around 10 years to develop, with different sources putting the average R&D spend (during the journey from the bench to the market) somewhere between just under $1 billion and over $2 billion. According to one study, at least a third of this cost is incurred during the drug discovery phase alone[1].

AI developers hope that, by applying their technology, they can rapidly identify promising and not-so-promising drug candidates before they enter clinical trials in a “fail fast and learn fast” approach. Some of these technologies are really now coming into their own. Exscientia – the company that has developed the first three AI-designed drug candidates to enter human clinical trials – boasts that its technology is helping to reduce the time taken, from target identification through to identifying a development drug candidate, by around 70% as compared to the industry average[2]. Recursion Pharmaceuticals also claims some impressive statistics – including that its development costs, from screening through to obtaining an investigational new drug application (IND) clearance for a drug candidate, are only around a quarter of the industry average[3].

AI tools also have a significant part to play in clinical trials, where it becomes increasingly costly for a drug to fail. By helping to optimise the design of trials and selecting those patients who are most likely to benefit from the treatment (rather than a “one-drug-fits-all” model), AI has the potential to improve drug approval rates and goes hand in hand with precision medicine.

Spectrum of commercial deals

So the key question is, what commercial models are players adopting in this field? We’re seeing a full spectrum of commercial deals – from AI developers using traditional service and licence fee models with their pharma customers at the one end, through to joint ventures and consortiums, and even pharma companies buying out AI companies outright at the other extreme. Some of the most high profile deals that have been announced in recent times lie in between these two poles, such as the collaboration between AstraZeneca and BenevolentAI. Since the partnership was first formed in 2019, two targets (one for chronic kidney disease, and the other idiopathic pulmonary fibrosis) have already been identified, validated and selected for AstraZeneca’s portfolio with the aid of BenevolentAI’s platform.

These types of strategic partnerships between big pharma and leading AI companies typically base themselves on the commercial model that has been tried and tested in drug discovery for many years. In this type of arrangement, the AI company (similar to the “traditional” drug discovery partner) initially receives an upfront sum or payment of its R&D costs and possibly, also, option fees as and when the pharma company selects which potential targets or molecules to take forward into the clinic. This is then followed by the AI company sharing in any downstream success in the form of lump sum payments that are linked to the pharma company achieving clinical, regulatory and/or commercial sales milestones, and/or royalty payments on any eventual sales. The payments (or, at least, the potential payments – your typical “biodollars”) can be considerable. To illustrate, in early January 2022, Exscientia concluded an agreement with Sanofi involving an upfront payment of $100 million, downstream milestones potentially worth up to $5.2 billion, plus tiered royalty payments[4]. Of course, not all deals based on this model have such staggering figures. We have also seen lesser known AI companies, whose AI tools are not as mature (and have perhaps not had the benefit of learning on other big pharma datasets), employing this commercial model but commanding much more modest amounts.

So far, so familiar – but of course not all deals follow this blueprint. Some of the AI companies that have been particularly successful in securing these types of deals are those that are (or are seeking to become) biotech companies in their own right. AI companies (such as those with their own drug development pipelines) that provide additional support for the R&D programmes of their collaboration partners (e.g. through using their own wet labs or working with their own CROs), at least initially until such time as the collaboration partner takes on responsibility for further pre-clinical and clinical studies, can more readily demonstrate their impact and therefore have a stronger argument to share in any subsequent success.

Many AI companies offer a traditional “fee for service” model. Increasingly, some of them are looking to move towards “outcomes-based” arrangements (sharing in the partner’s revenues and/or cost savings – e.g. in a clinical setting), but so far with varying degrees of success. Making the breakthrough is not always straightforward. Some of the more established AI companies that we have spoken to believe that the key lies in generating and securing a stake in the arising IP, and building value in and around the AI platform – at the heart of which, lies the data.

Data – the key differentiator

For many pharma and biotech companies looking to partner with an AI company, it is not necessarily its AI tools that are the differentiator, but the datasets to which the AI company has access – i.e. the data is the new oil. AI providers that have been able to build up, cleanse and transform, and build models from, large quality datasets from a diverse range of sources (including private sources as well as public) are those best positioned to negotiate a financial “upside” if any of the discoveries are taken forward by the partner.

So what about organisations that provide data to AI companies in order to train the AI in the first place – how do they generate value from their contributions? Holders of rich sources of data, including (but certainly by no means limited to) public bodies such as biobanks, healthcare providers and research institutes, can often struggle to monetise their data when making it available for use in an AI project. The difficulty lies in trying to justify and put a value on datasets which may or may not assist in the development of a number of different commercial products or services (and which often are not necessarily known or envisaged the time of contracting) and which in any event are several steps from the end result. However, this does not mean that it is not possible (or indeed appropriate) in some cases to get value recognition in cases where the data in question is likely to be transformative in training or improving the other party’s AI system. This recognition could, for example, take the form of royalties (although perhaps limited in time and/or amount) or, possibly, even a one off fee if the AI company is listed or sold for a significant price shortly afterwards. Sensyne Health, for example, has concluded a number of strategic research agreements with NHS trusts who have contributed data to build up Sensyne’s AI tools and who, in return, have been given an equity stake in the company (amongst other benefits).

Some of the more innovative solutions we have seen include the marketplace model operated by Eagle Genomics (among others). This allows customers who have contributed data to its AI-powered microbiome platform to share in some of the revenues generated from it or otherwise to benefit from reduced access fees – essentially, allowing customers to monetise what might be a niche dataset that has little value on its own.

M&A landscape and in-house AI programmes – will big pharma take over?

Some life sciences companies have already snapped up AI or ML technology companies, such as Roche’s acquisition of Flatiron Health in 2018. However, more frequently, pharma companies are making strategic investments in AI companies. Sometimes the investment is made alongside a collaboration deal between the two parties, such as when Bayer led the investment round in Recursion Pharmaceuticals in 2020. In other cases we have seen the investment made earlier, perhaps with a view to a possible collaboration later down the line as the AI platform evolves, or later (after signing a collaboration agreement) when it becomes evident to the pharma company that there is a prospect of making a healthy return on an investment. In December 2021, it was announced that BenevolentAI signed a deal to be taken public via the special-purpose acquisition company (SPAC), Odyssey Acquisition, with a post money valuation at €1.5 billion – Europe’s biggest ever healthcare SPAC merger – in which its strategic partner, AstraZeneca, also has a stake.

One of the issues facing pharma and biotech companies when collaborating with an AI company is the attraction of working with AI that has already learned from the datasets of other similar companies. However, by providing data to the AI company, the pharma or biotech company may well improve the AI company’s tools, algorithms and models for the benefit of its competitors – unsurprisingly, an unattractive proposition. In some cases, an AI provider will work with a pharma company using a separate instance of its platform, either creating tools specifically for that customer or sometimes agreeing to a moratorium on feeding the developments back into its core system for the benefit of its other clients – although these types of arrangements tend to be much less common.

Some major pharmaceutical companies (including GSK) have been making significant strides towards building their own AI capability in-house, as this allows them to have greater control over the AI tools, datasets and other assets – and, importantly, avoids the thorny issue of enabling competitors. One of the main challenges of “going it alone” is how to attract the right talent, given how difficult it can be to find individuals with the necessary skill sets in the crosstab of AI and biology – although one suspects that today’s graduates will be able to better plug this gap in the future. At London’s Genesis Life Sciences conference in December 2021, some opinion leaders were already predicting that the AI providers’ bubble will burst before too long, with other major pharmaceutical companies most likely to follow suit and take more AI programs in-house. We will watch these developments with great interest.

[2] (pg 141)
[3], “Decoding Biology to Radically Improve Lives”, End of Qtr 3, 2021 (p13).