SinS 2025: AZ outlines its digital transformation of analytical science to accelerate drug discovery pipeline

At the Solutions in Science 2025 conference in Brighton, United Kingdom, Dr Nichola Davies, Director of Analytical and Structural Chemistry within the Oncology Research and Development group at AstraZeneca (AZ), gave a presentation of how AZ has transformed its analytical science infrastructure to support a more predictive, scalable and efficient approach to drug discovery.

Davies described how her team has embedded FAIR – Findble, Accessible, Interoperable and Reusable – data principles into a digitally connected laboratory environment. This integration has enabled AZ to address longstanding operational inefficiencies, improve levels of reproducibility and to apply advanced machine learning tools in analytical workflows.

“Historically, our industry has approached scientific challenges sequentially,” said Davies.

“We now aim to interrogate data holistically to reduce the number of iterations in the design–make–test–analyse cycle and improve candidate [drug] quality.”

At the heart of this transformation has been a strategic overhaul of AstraZeneca’s data architecture in support of the ‘design–make–test–analyse’ cycle used in drug development. A central cloud-based platform now integrates experimental data and metadata from diverse analytical technologies including mass spectrometry, nuclear magnetic resonance (NMR) spectroscopy, purification processes and screening outputs.

This infrastructure has enabled chemists to retrieve related datasets that were previously siloed, accelerating hypothesis generation and experimental planning. Davies explained that experimental metadata now captures a broad range of contextual information, from instrument configuration and analyst identifiers to the scientific purpose of the analysis.

In daily workflows, scientists use barcoded sample vials and an integrated logistics and scheduling system – pneumatic tube transport systems – to deliver samples to the appropriate analytical method and, critically the next available piece of analytical tool, not  just the one nearest to the door to the lab. A critical enabler of this system has been real-time instrumentation monitoring. Davies outlined how the company has implemented automated dashboards to track analytical performance using a colour-coded traffic light system to flag deviations. These quality control tools have facilitated proactive maintenance and improved reproducibility across high-volume operations.

Results are returned to the scientist’s digital workspace, linked directly to compound registries and experimental records.

Machine learning models are now embedded throughout AZ’s analytical environment. Davies described how chromatography prediction tools, trained on historical data, anticipate retention times and suggest optimal conditions for compounds under investigation. This predictive capability allows chemists to avoid analytically intractable designs at an early stage, improving efficiency and reducing failure rates.

“Models can now warn if a proposed molecule is likely to precipitate under standard liquid chromatography conditions enabling chemists to refine synthetic plans before a single experiment is run,” said Davies.

In a further case study, Davies explained how AZ has automated the structural verification of synthesised compounds. By integrating synthesis planning software with spectroscopic analysis, the system compares predicted and observed spectra to rank candidate structures. This approach has improved both the speed and reliability of compound verification.

To increase confidence in these assessments, her team has also combined orthogonal techniques such as infrared spectroscopy and NMR. This multi-modal approach has proven valuable for structurally similar analogues or stereoisomers, which are difficult to resolve with the application of a single technique.

Although structure elucidation remains a complex task, AZ is already piloting transformer-based models that aim to automate this process entirely. Davies noted that while such systems are still in development, they offer the potential to establish closed-loop feedback between compound design and structural verification.

Davies concluded by emphasising that the system developed by AZ is not a generic data lake but a context-aware infrastructure tailored to the needs of pharmaceutical R&D. The platform has already ingested hundreds of thousands of experiments and underpins a wide range of activities, from early discovery through the pipeline to the endpoint of regulatory submissions.

By improving data discoverability, supporting predictive modelling, and enhancing reproducibility, the platform has helped AZ to accelerate timelines, reduce resource consumption and improve the organisation’s decision-making. Davies acknowledged that this transformation is still a work in progress but characterised it as a critical step toward a harmonised, digitally enabled and machine-learning assisted model for effective scientific research.