Written by:
Vice President, Data Science & AI, AstraZeneca
Chief Data Scientist Advisor, AstraZeneca; Professor of Public Health & Clinical Informatics and Associate Pro Vice Chancellor for Innovation, University of Liverpool
Our scientists are using data science and artificial intelligence (AI) to engineer new and better ways to discover and test the potential medicines of tomorrow, and to connect them to care. Here, we examine some examples for how we are applying data science and AI in the discovery and development of the potential next wave of innovative medicines.
AI and knowledge graphs for drug discovery
Drug discovery has evolved through biochemical, biological, and most recently, biotechnical experimentation. In 2012, Google introduced the term ‘Knowledge Graph’, drawing from previous decades of computer science – to depict how knowledge can be organised, represented and reasoned. Knowledge Graphs bring incredible libraries of information to life, helping users spot the connections between thousands of different sources.
For drug discovery, we are using Knowledge Graphs to harness vast networks of scientific data to give our scientists the information they need about genes, proteins, diseases and drugs, and their relationships – how they interact, work together, or work against each other. Having a better understanding of these relationships can assist research teams in finding important connections across the latest web of biological mechanisms and candidate medicines. For example, we are now using disease-specific Knowledge Graphs to better understand complex multi-factorial diseases such as idiopathic pulmonary fibrosis and chronic kidney disease, in partnership with BenevolentAI.
Such networks can also be generated through emerging AI-driven language models, such as ChatGPT, GatorTron and ClinicalBERT – technologies that could enable the analysis of vast amounts of clinical text and scientific literature. Importantly, use of such types of models has the potential to help scale up the continuous improvement and repurposing of medicines for new disease indications.
Data science to improve clinical practice and clinical trials
Advances in data science and data utilisation are key to improving clinical trials and real-world evidence through which medicines are regulated and optimised. The quest for inclusive trials seeks fair representation of patient groups across all lived-experiences – including people from low- and middle-income countries or communities – who may be more likely to have an earlier onset of a wide range of medical conditions and be more at risk of having multiple long-term conditions.
Electronic health records are a key data source for better understanding patients, both individually and within populations. As electronic health record data are better captured, linked and curated, opportunities to understand disease risks and trajectories improve.1-3 Such data also allow for optimised data processing that could be reused to improve clinical trials’ feasibility analyses, recruitment, safety surveillance, economic evaluations, generalisability studies and long-term outcomes surveillance.
Recent advances in causal machine learning have the potential to enhance patient care, public health measures, service quality management, planning and research – including for clinical trials. For example, machine learning approaches such as Convolutional Neural Networks (CNNs), Long Short-Term Memory networks (LSTMs), and Generative Adversarial Networks (GANs) are starting to enable innovations such as the estimation of treatment effects or the generation of synthetically balanced case-control populations and ‘virtual control groups’. By using machine learning methods (and using data from collected from past clinical trials, natural history studies, electronic health records, claims data, or disease registries) to create virtual control groups, we can move more of our study designs away from placebo control arms. With less dependence on human controls, more participants receive the innovative treatment rather than a placebo or standard care.
Programmes not just prescriptions
Personal health information is evolving from passive clinical records into interactive combinations of records, data-streams and algorithms. In many parts of the world, the public are coming to expect more self-service access to healthcare and interaction with human or AI-driven services, between their in-person contacts with clinicians – increasingly reflecting how people live their daily lives in a connected world. For example, phone apps that integrate and enhance care are becoming more and more common in regular clinical practice as well as part of clinical trials. Furthermore, it is theoretically possible to produce a companion AI for medicines that augments drug regimens (e.g. personalised through pharmacogenomics), supports medication experiences and provides rich feedback to researchers and regulators.
To meet the needs of patients, we are aiming to deliver frictionless clinical trial experiences that integrate seamlessly into participants’ lives during the trial and beyond. One way we are enabling this is through Unify. Unify is an app that brings together a range of apps, websites and devices into an easy-to-use single tool to enable data collection, streamline the experience for trial sites and support patients during the trial and beyond. Unify is designed to foster the connection between the physician and patient and enhance the effectiveness of their relationship to achieve the best clinical trial experience possible.
A fully connected medicine can be thought of as a therapeutic package of agent and AI. For optimal patient, provider and social outcomes from medicines, these technologies and wider frameworks will need to be adopted globally. Though this presents inherent challenges, we are confident that we can make progress by listening to and learning from patients while continuing to innovate with our digital and AI-led technologies that bring us closer to connected medicines.
Data science and AI at AstraZeneca
At AstraZeneca, we harness data and technology to accelerate the delivery of potential new medicines, innovate to drive efficiency and success, and pioneer the tools and techniques that maintain our competitive edge. We are embedding data science and AI across all of our R&D activities, from target identification to clinical trials, to identify new opportunities to push the boundaries of science to deliver life-changing medicines.
