How are AI and multiomic data improving drug safety and efficacy prediction?

Written by:

Jorrit Hornberg profile

Jorrit Hornberg

Vice President, Global Head of Safety Sciences, Clinical Pharmacology & Safety Sciences, R&D, AstraZeneca

Richard Goodwin profile

Richard Goodwin

Global Head of Predictive Data Sciences, Clinical Pharmacology & Safety Sciences, R&D, AstraZeneca


Predicting how a potential medicine will interact with human biology, to evaluate its safety and efficacy profile, is one of the most important challenges in preclinical drug development. At AstraZeneca, multiomics and advanced cell models are combined with AI drug development approaches to help us predict this with greater human relevance, than was previously possible.


At AstraZeneca, our NAMs capability is an integrated set of innovative technologies, including advanced cell models, multiomics and AI.


Why is predicting drug safety and efficacy so important in drug discovery?

Every drug candidate that reaches a patient has gone through rigorous safety and efficacy evaluation before it enters the clinic. Multiomic data and AI, part of our New Approach Methodologies (NAMs) capability, are helping us decide which drug candidates to advance to clinical development by enabling us to:

  • Understand how a drug candidate interacts with human biology in greater depth.
  • Identify the most promising drug candidates faster and earlier in drug discovery, supporting more informed candidate selection.
  • Support better-informed decisions about which candidates to progress.
  • Increase the probability of clinical success and potential patient benefit.
  • Connect insights across the drug discovery lifecycle, from early research to the clinic.

What is multiomics?

Multiomics, or multiomic analysis, is the study of many layers of molecular information, including genes, RNA, and proteins, evaluated together. It gives scientists a rich, multi-layered picture that places drug candidate data in the context of the wider biological landscape, rather than a single snapshot.

How are AI and multiomic data used in drug discovery?

Multiomics provides a multi-layered view of biology by combining information from different sources, such as transcriptomic and proteomic data from advanced cell models or clinical data, and pharmacokinetic and pharmacodynamic (PK/PD) and digital pathology data. AI helps integrate and interpret these complex datasets, identifying patterns and generating more predictive, actionable insights than any individual data type could provide alone. This data-driven approach to drug discovery supports earlier biomarker discovery and more informed candidate selection.


At AstraZeneca, we apply predictive models built on multiomic data to drug safety and efficacy evaluation across organ systems.


How are AI and proteomics helping improve drug candidate design?

For a type of therapeutic modality called targeted protein degraders, our scientists combined high-throughput proteomics with machine learning to identify specific biological signals early in discovery. The approach helped scientists to identify improved version of the molecule that preserved its therapeutic activity while achieving a more favourable safety profile.



Connecting multiomic data to AI is fundamental to how we advance predictive science. Each dataset compounds AI's predictive value, supporting better-informed decisions about the drug candidates we progress.

Richard Goodwin Global Head of Predictive Data Sciences, Clinical Pharmacology & Safety Sciences, R&D, AstraZeneca

How do insights from patients inform early drug discovery?

Predictive science is at its most valuable when insight flows in both directions: from early research toward the clinic, and from clinical evidence back into how we design and select new drug candidates. This is known as back translation, and it is a well-established part of how we work.

AI models trained on preclinical multiomic data are tested and refined against clinical and real-world evidence, so predictions made at the earliest stages of discovery become progressively more relevant to patients. AI is also being used to identify biomarkers from routine pathology images that help clinicians select the patients most likely to benefit, an example of predictive science already informing decisions in the clinic.


NAMs combine advanced cell models, with multiomics data, quantitative modelling and AI trained on historical laboratory and clinical data.


What is back translation?

Back translation is when insights from clinical data, how medicines behave in patients, are fed back into earlier research. It helps make future predictions more accurate and more relevant to real patient outcomes.



How is agentic AI supporting drug discovery at AstraZeneca?

We are increasingly using agentic AI to automate and connect the many steps involved in processing and reusing large-scale scientific data, for example running analyses, applying predictive models, and preparing data so that insights generated for one question can be reused for the next.

NAMs work best when integrating large amounts of data to answer specific, well-defined questions.


What is agentic AI?

Agentic AI systems can coordinate multiple steps of a scientific workflow, for example running an analysis, checking the result, and deciding what to do next, within clearly defined boundaries set by scientists. The predictive insight itself comes from the underlying interconnected data, which include foundation models. Foundation models are large, domain specific AI models trained on vast datasets of multiomic data that can be adapted to many scientific questions.



By making data and insights reusable across programmes, agentic AI helps us derive greater value from every experiment, accelerating how quickly we can move from a scientific question to an actionable answer.

Kevin Moreau Senior Director, Safety Omics, Clinical Pharmacology & Safety Sciences, R&D, AstraZeneca

How does AstraZeneca ensure AI in drug discovery is trustworthy?

AI-derived insights that inform how a drug candidate will interact with human biology must be responsible and explainable. That is why AstraZeneca's approach to AI is fit-for-purpose and ethically designed:

  • Models are built for defined scientific questions, and predictions are interpretable, so scientists understand why a model reaches a conclusion.
  • Human expertise remains central to every decision.
  • Data is protected through strict security, access controls, and patient privacy safeguards, and applications are aligned with regulatory expectations.
  • In a highly regulated environment, this approach supports innovation and allows the science to be trusted by regulators, clinicians, and ultimately patients.

What is the future of AI in drug safety and efficacy prediction at AstraZeneca?

As these capabilities mature, we are moving from AI-enabled predictive science toward AI-first predictive science. This takes us from AI supporting human decisions to where AI generates and tests hypotheses about safety and efficacy that our experts then validate and refine. The goal is not to replace scientific judgement, but to enhance it: designing better molecules, making earlier and more informed decisions, and bringing the right medicines to the right patients faster. Where these approaches replace resource-intensive legacy processes, they also have the potential to make R&D more sustainable over time.


AI and multiomics are transforming how we evaluate potential new medicines, turning complex data into actionable insight that supports more confident decisions across our pipeline and our goal to deliver better outcomes for patients.

Jurre Kamphorst Director of Multiomics, Integrated Bioanalysis, Clinical Pharmacology & Safety Sciences, R&D, AstraZeneca

Topic:
 




Veeva ID: Z4-90455
Date of Preparation: November 2026