As long as drug discovery, development and commercialisation activities and budgets continue to be managed in isolation, and Generative AI projects applied in a piecemeal fashion to individual applications, the pharma industry risks inhibiting the technology’s broader potential. Daniel Jamieson, CEO & founder of Biorelate, sets out his vision for a more strategic approach to generative AI as an enterprise-wide R&D assistant, and its potential in honing new drug success.
In life sciences R&D there are two main strategies for using generative artificial intelligence (AI), the deep-learning algorithms that can distil complex knowledge into easily digestible summaries. Deploying the technology within individual departments makes sense for many use cases. However, there is a bigger opportunity which is being exploited far less. That is to use Gen AI across all departments to bring a company’s entire R&D strategy together in a super decision intelligence tool, and there are steps companies could be taking to position themselves for this broader opportunity which is just over the horizon.
So why might this be important?
Factoring in the relatively high failure rates, the average total expenditure to introduce a novel drug to market ranges from $2.5 billion to over $6 billion. This figure is experiencing a consistent upward trend, parallel to the rise in failure rates as biopharma companies grow more ambitious in their product and therapy innovation.
While there may be no shortage of new drug ideas, selecting candidates with the greatest likelihood of success remains a painful and costly challenge. The opportunity is to deduce which new drugs are worth pursuing based on everything important that is already known about that field. Targeted use of modern AI can help uncover unprecedented and even very subtle intelligence that might otherwise be overlooked. Together these insights offer a new level of confidence in shortlisted candidates, while also making it possible to eliminate other prospects earlier in the process which may have looked promising initially.
This strategic approach could enable significant cost savings in the broader drug development procedure, and importantly, expedite the delivery of more drugs to patients who need them most.
Defining generative AI
Generative AI is a subset of artificial intelligence that, through machine learning, is able to very rapidly distil key information and insights from vast knowledge banks, and create new output from it, which is intuitive and easy to digest. In life sciences, as well as being used to draft articles, the technology can assist researchers in generating scientific content or summaries.
For generative AI to be trusted as a reliable ‘source of truth’ in life sciences applications, knowledge sources need be both transparent and validated, along with how connections have been made. This is being addressed via approaches to data curation and quality control such as Retrieval-Augmented Generation (RAG).
Where it is applied appropriately and thoughtfully, generative AI makes it possible to harness smart, mass-scale data analytics in a reliable and accessible way in drug discovery. And the more that applications are connected and driven by an overarching strategy, the greater the potential returns.
Solving drug discovery data challenges
No drug developer can afford to waste time and resources repeating studies that have already been conducted, or pursuing problematic next-generation medicines – unless they can pinpoint the insights that will make a critical difference to success.
The problem is how to get to those insights reliably and efficiently. Published research and other texts may contain rich knowledge, but for the most part this remains un-curated and of a scale that feels impenetrable. It is thought that as much as 80% of healthcare data today is inaccessible for analysis. Up to now, data has had to be largely manually curated, consuming considerable resources and carrying the risk of something important being missed. Although cause-and-effect-like relationships are abundant in the available text, they remain difficult to find, connect and analyse.
Targeted use of AI makes it possible to capture and link findings so discovery teams can infer new meaning and insights. In one scientific study, researchers may have examined how a specific drug triggers the activation of a particular protein. Concurrently, another separate study may highlight that this activated protein is linked to the onset of hypertension. While individually these findings provide valuable insights, it is only when they are connected that a potential hypothesis emerges: in this case, that the drug in question might pose a risk for inducing hypertension.
The powerful enabler, harnessed by generative AI, is the ability to structure and analyse valuable data that up to now has been locked in text. This could be from internal archives and/or from millions of scientific research articles, in combination with all the other structured data sources, such as transcriptomics and proteomics.
Every drug discovery company needs to optimise spending and control costs along the development cycle. Anything that can help organisations fail faster and improve the speed to approval and market access, by reducing the need for additional experiments, helps towards that goal.
At a macro scale across potentially hundreds of drug programmes, even just a 1-2% improvement in a drug’s chances of success, accelerated market delivery, or the chance to capitalise on additional opportunities, can have a huge impact across an entire drug discovery pipeline.
Targeted, appropriate application of AI to elicit valuable insights from across vast research archives can yield 10- to 100-fold improvements at certain parts of the clinical trials process. Expediting early target selection to Phase I clinical trials from what might have been four and a half years to just one represents a significant improvement in speed.
Optimising the AI opportunity: The importance of strategy
Fragmented operations and budgets can prevent drug developers from capitalising on the broader benefits of AI-enabled insights. Up to now it has been rare to see a single business-level objective that transcends R&D and extends through to commercial planning in a way that can influence everything in an integrated way.
Generative AI’s opportunity here is to serve as a kind of ‘chatbot’ or knowledge assistant for the organisation. Once it’s possible to link data from R&D, the commercial side and market and marketing data, even if each function or therapy area is running its own strategy and programmes, it becomes possible to access and apply guidance based on the full picture of what’s going on.
GSK is already making good headway here. It has been creating its own LLMs for specialist tasks, even creating its own proprietary LLM-based operating system. Its conversational interface allows users to explore complex research questions without needing a detailed understanding of the company’s data ecosystem. GSK staff can automatically discover and extract relevant data from structured and unstructured sources (e.g. databases and literature), analyse and reason about that data, and synthesise the resulting insights into an actionable conclusion.
Making AI pay in drug discovery: Next steps
Generative AI offers a new way to interface with data and other AI models (‘agents’). Those AI agents might each have very specific roles, but they also contribute to the bigger picture, allowing users to ask not only about the financial performance of biopharma companies and their drugs, but more probingly: “What are the best-performing drugs on the market, and what are their common mechanism of actions?” (Here, the meta model queries to deliver the answer.) Companies like GSK are already experimenting with the potential.
AI continues to evolve of course, too, adding to the possibilities. In 2024, multi-modal algorithms (MLMs) will become a large and growing trend, presenting the opportunity for teams to interrogate not only text, but also images, sound and video.
More important than the potential applications however, is the robustness and reliability of the AI capability. Unless teams can absolutely trust the validity of the findings that are returned, and trace these back to their source, further painstaking assessments and analyses will always be needed. It is here, then, that companies need to be at least as focused and discerning in their technology and deployment choices.
About the author
Daniel Jamieson is CEO of Biorelate, which enables curation and smart linking of currently digitally unsearchable materials, and external archives of biomedical knowledge, to accelerate new drug discovery. Via its platform, Galactic AI, Biorelate uses AI-led curation to provide and enable the insights that matter to scientists and organisations developing the innovations of the future, helping them to advance and expedite the delivery of biomedical breakthroughs.