Data, education, and progress: strategies for GenAI adoption in drug development

Generative AI (GenAI) has entered the world of drug development, reshaping its landscape through tools that can accelerate discovery, optimize processes, and improve outcomes. However, GenAI’s successful integration requires more than just technological readiness—it demands robust data strategies, cross-disciplinary collaboration, and a culture of learning and adaptation. 

In a recent panel discussion at BioTechX 2024 in Basel, experts from Novo Nordisk, AstraZeneca, and academia came together to explore a few key considerations relative to the adoption of GenAI for drug development. Moderated by ONTOFORCE’s Bérénice Wulbrecht, the panel explores the role of data as GenAI’s fuel, the challenges and opportunities GenAI introduces, the human factor in adoption and trust, and the practical steps needed to integrate it effectively into pipelines. This blog delves into these topics, weaving in insights and anecdotes from the panel’s experts 

Data as the fuel for GenAI 

GenAI, by design, is data-hungry. Its capabilities are only as strong as the data it consumes. According to Michel Dumontier, co-author of the FAIR (findable, accessible, interoperable, reusable) data principles, preparing data to be AI-ready is not just a best practice but a necessity. For GenAI to generate meaningful insights, data must be properly structured and interoperable, in this way, the FAIR principles and GenAI are truly made for each other.  

Mathew Woodwark, Head of Data Standards and Interoperability at AstraZeneca, shared a pragmatic take: data structuring needs to match the specific use case. For general analysis such as a summarization of information, less rigor may suffice. However, when AI is tasked with something as precise as determining dosage ranges, highly curated data becomes critical. Without this foundation, even the most advanced AI models risk producing unreliable results—garbage in, garbage out. 

Challenges and opportunities in drug development 

The application of GenAI in drug development presents a paradox of immense potential and significant hurdles. Allan Christian Shaw, Corporate Vice President of Data & Knowledge Discovery at Novo Nordisk, pointed to a trend where AI is evolving from a mere assistant to an autonomous optimizer in specific areas, like closed-loop experimentation. Here, AI determines the next steps in research, speeding up processes that previously took weeks or months.  

However, as Dumontier noted, GenAI tools are fundamentally predictive rather than deterministic. While they excel at generating human-like responses or summarizing data, they can sometimes hallucinate—creating outputs that seem plausible but are incorrect. This makes their use in critical decision-making, such as regulatory submissions or clinical trials, a cautious endeavor. 

Woodwark’s experience at AstraZeneca echoed this caution. Regulatory environments demand explainability and accountability. AI systems cannot operate in a black box; every decision must be traceable back to validated data and methodologies. This is particularly crucial when using AI to guide patient treatments or optimizing molecule development. A human-in-the-loop approach is thus necessary.  

Yet, the opportunities are undeniable. AI-powered knowledge graphs and natural language processing tools are already enabling researchers to uncover insights from vast datasets, accelerating lead discovery and decision-making. These tools, when used as decision-support systems, can significantly enhance productivity and innovation. 

User adoption, education, and trust in GenAI 

The promise of Gen AI is only as impactful as its acceptance by the people using it. Adoption requires not just enthusiasm but trust, education, and clear expectations. 

Woodwark observed a surge of excitement among teams exploring Gen AI’s capabilities. However, this enthusiasm often clashes with a lack of understanding about the tool's limitations. Early experiences with tools like ChatGPT revealed its tendency to hallucinate, which initially dampened trust among scientific users at Novo Nordisk. For Gen AI to gain widespread acceptance in science-driven organizations, users need confidence in its reliability. 

Education is the linchpin here. Dumontier emphasized that education around AI’s strengths and limitations must extend beyond just data scientists to reach chemists, biologists, regulatory professionals, and other profiles in the field of life sciences. AstraZeneca, for instance, has developed comprehensive educational resources to help employees understand Gen AI’s role and how to leverage it responsibly. 

Gen AI’s democratization has also reshaped collaboration patterns. Traditional workflows often separated domain experts from data scientists. Now, natural language interfaces are allowing domain experts to query complex datasets directly. As Woodwark noted, this shift necessitates changes in how AI/data teams engage with their stakeholders, creating a new dynamic of real-time problem-solving. 

Practical advice for integrating GenAI into drug development pipelines 

For organizations seeking to embrace Gen AI, the panelists offered practical advice grounded in their experiences: 

1. Start small: Shaw recommended starting with specific, manageable use cases that demonstrate clear value. Early wins, such as automating image analysis or accelerating data retrieval, can build confidence and momentum within the organization.
2. Invest in data preparation: Woodwark stressed that well-structured and labeled data is the foundation for meaningful AI insights. Organizations should prioritize building robust data pipelines before scaling AI solutions.
3. Foster collaboration: Dumontier advocated for hackathons and proof-of-concept projects as a way to bring teams together, test ideas, and spark innovation. However, he also cautioned against stopping at the proof-of-concept stage. Go the extra mile: demonstrate that this concept could work on a broader set of inputs and importantly identify limitations to avoid disappointment later.
4. Manage expectations: Both Shaw and Woodwark highlighted the importance of setting realistic expectations about what GenAI can achieve. Educating users on its capabilities and limitations can prevent disillusionment and ensure sustained engagement. 

Learn more: watch the full panel discussion 

The journey to realizing GenAI’s full potential is as much about people as it is about technology. By focusing on a strong data foundation, education, and incremental progress, organizations can harness this transformative technology to its fullest, unlocking breakthroughs that could redefine the future of healthcare. 

The panelists shared a wealth of insights and real-world anecdotes. Watch the full discussion to hear about their experiences and practical advice in greater depth. Watch now >>>