Agentic AI is closing the gap between data and wet-lab scientists
One of the slow leaks in early drug discovery has been the split between wet-lab scientists and data scientists. The disciplines depend on each other, but tools, workflows, and language often keep them apart.
Dr. Natalie van Zuydam, Executive Director of Data Sciences at AstraZeneca, is tackling this head-on. Speaking in Liverpool at ELRIG Drug Discovery 2025, she laid out a practical path for using agentic AI and predictive models to make advanced analytics usable by every scientist-and to give time back to the bench.
From the bench to data science to pharma
Dr. van Zuydam started in the wet lab in South Africa, earning a Master's in molecular biology at the University of Pretoria with a focus on plant pathogens and tree health. A move to London led to an analyst role at St George's Hospital in a stroke genetics group, where collecting patient samples created a direct link between data and lives. That connection pushed her to pursue a PhD at the University of Dundee studying macrovascular and microvascular complications in type II diabetes.
Postdocs followed at the University of Oxford with Prof. Mark McCarthy, where the team generated both data and new analytics. That experience set the stage for a transition into pharma-motivated by the chance that work done at the earliest stages might one day improve a patient's life.
Three practical layers of AI for drug discovery
- Layer 1: General LLMs - Tools like ChatGPT and co-pilots help with meeting notes, drafting documents, and clearing admin noise. They let scientists spend more time thinking and discussing, less time formatting.
- Layer 2: Life science assistants - Bespoke tools that link automation, handle machine booking, and act like virtual research assistants. They free both data and wet-lab scientists to focus on the work only they can do.
- Layer 3: Predictive models - Machine learning that can replace parts of an experiment. The goal: smaller, more focused studies that move projects faster through the pipeline.
What agentic AI actually changes
Agentic AI sits across these layers and removes friction. Pair a model like AlphaFold with a natural-language agent, and a wet-lab scientist can run complex workflows without deep data science expertise. The barrier drops. Usage climbs. Feedback loops tighten.
Some worry this could reduce interaction. Dr. van Zuydam sees the opposite: more shared context, faster iteration, and better handoffs. Collaboration gets easier when everyone can access the same models with plain language.
Publishing, partnerships, and patient focus
AstraZeneca continues to publish and collaborate with academia, opening access to bespoke omic datasets at a scale that used to be out of reach. The next step is turning that complexity into decisions with AI and machine learning.
Underpinning it all is a patient-first mindset. That clarity makes partnerships simpler and goals aligned. Still, there's responsibility here-use the tech wisely, be honest about where it adds value, and accept that it won't answer every question.
How to apply this in your lab
- Start with LLMs for routine work: meeting notes, SOP drafts, experiment summaries, and protocol comparisons.
- Connect agents to the tools you already use: automation scripts, booking systems, LIMS, ELNs, and data catalogs.
- Pilot a predictive model to replace a slice of a workflow (e.g., triage candidates or reduce assay runs). Measure time and quality gains.
- Make models usable via natural language and templates. Shorten the path from question to output.
- Stand up a cross-functional review: wet-lab, data science, IT, and compliance. Agree on validation criteria and guardrails.
- Publish where you can and set up academic ties for specialized omics. Shared data creates shared progress.
Bottom line
Agentic AI helps scientists spend less time wrangling tools and more time doing science. The wall between the bench and the model is getting thinner-and that's good for teams, timelines, and patients.
If you're upskilling your team on practical AI for research, explore role-specific learning paths here: AI courses by job.
Your membership also unlocks:
