ORIGINALLY PUBLISHED
11 July 2024
Written by:
Vice President of Biometrics, Oncology R&D, AstraZeneca
Executive Director, Statistical Programming, AstraZeneca
Executive Director, Real World Evidence, AstraZeneca
The global impact of digital transformation in the last few decades cannot be understated. And now, generative artificial intelligence (AI) is making headlines for its potential to transform lives. By applying this rapidly evolving technology in the world of drug discovery, we are equipped to take our capabilities to the next level with one goal in mind — improving patients’ lives.
Leveraging innovative data science and generative AI in oncology R&D
Innovative data science tools have been embedded across our R&D process for years. We’ve harnessed transformational technologies to analyse data, accelerate clinical trials, gain a better understanding of new diseases and more, all to enhance delivery of new medicines for patients.
Recent advances in tools like generative AI and open-source programming languages are now creating opportunities to further push our capabilities and the boundaries of science. This of course requires a deep understanding of the benefits and potential challenges of using the tools available for use to ensure we are maximising our impact.
Generative AI and oncology biometrics
Generative AI describes models or algorithms that use natural language processing and machine learning to process enormous amounts of existing data and generate new content, such as text, images or even music.1 At AstraZeneca, we are now leveraging tools that use generative AI as we pursue scientific innovation to identify new targets and explore novel treatments for the greatest potential benefit to patients. Although there is some hesitation around the use of generative AI in some applications, as with any new or evolving technology, we see and are embracing its great potential in drug development.
In R&D, generative AI has potential to:
- Assimilate complex data and evolving disease information to project outcome scenarios for treatment decisions.2
- Predict what molecules and dosage regimens are potentially tolerable and efficacious to combine, using complicated algorithms.3
- Inform clinical trial design using real-world evidence (RWE) and complex algorithms to improve outcomes.4
- Transform productivity across teams by automating processes that would have typically taken hours.4
For example, in statistical programming, generative AI can be useful in alleviating administrative burden by automatising routine tasks and streamlining project management. It also aids in the development of more robust coding –– reducing coding errors and better detecting data anomalies.
There are also incredible opportunities for generative AI to boost efficiency in the workplace, such as in developing the introduction of study protocols for clinical teams, which could save up to 25% of team members’ time.4
Generative AI holds immense potential to take our R&D process to the next level.
Already we are seeing the benefits of generative AI from identifying novel targets to more efficient design of small and large molecules to informing clinical trial design and improving efficiency of our regulatory submissions.
Improving clinical trials and the capture and analysis of real-world evidence
The way we design and conduct clinical trials can benefit from RWE, which are the findings, insights and conclusions derived from the analysis of real-world data (RWD). RWD relate to patient health status or the delivery of health care and are routinely collected from a variety of sources. As this broad definition evolves, it has expanded to data being generated by patients in real-time through their wearable devices and ultimately their consumer digital health devices.
The veracity of all these data being generated and the irrelevant information, or noise, that comes with it, makes it challenging and time-consuming to glean useful information and ultimately evidence. By utilising generative AI, we’re able to train and fine-tune foundational models that can identify critical signals from the noise, enabling researchers to review important takeaways and generate hypotheses that can be validated through more traditional research methods.
The process of capturing RWE has matured rapidly to enable collecting real-time and relevant information for both healthcare professionals and patients to make more informed decisions. These developments in combination with generative AI raise several exciting possibilities. With patient parameters in place, such as medical history and genetic data, a physician could input this data alongside RWE and receive a personalised summary in real-time of the latest treatment options available based on the patient’s unique profile. Additionally, generative AI could inform physicians of relevant new clinical trial opportunities they were unaware of or didn’t have time to explore during busy working hours.
Analysing data in new ways with R
Data science ultimately is a science, and to have an impact on the public it must be translated into something digestible for other audiences. R is a powerful, open-source tool we can use to analyse data and share it in a beautiful interface – making data accessible to non-statisticians.
Using R has enormous advantages in data science. It provides a comprehensive toolset for data visualisation, bioinformatics, clinical trial analysis and more. R can quickly analyse swaths of data (including RWE or genomics) and produce exploratory analyses, enabling more efficient decision making when evaluating say, a clinical trial’s efficiency. When R’s ability to process large datasets is combined with generative AI, creative opportunities abound. Researchers can produce high-quality graphics through R Shiny, for example, and we can evaluate Clinical trial scenarios before they’re run to optimise trial designs and predict potential outcomes for patients.
Advancing oncology R&D strategy to the next level
Beyond capturing and leveraging patient data, new data science tools like generative AI and R may help us optimise the treatment options we choose to study and advance toward the market. Throughout the past decade, we’ve seen an incredible push to advance combination treatments in oncology.5 These have shown promise in improving response rates, delaying disease progression and increasing overall survival in some indications.5
However, the process by which we identify combinations is never easy and optimising suitable combinations can be a lengthy process.6 By analysing existing information on molecules and their mechanisms, generative AI can help us assess the safety profile of combining certain molecules and identify which ones could be more efficacious when combined. Currently, much of this is tested iteratively through platform trials, where multiple interventions are compared to the same control group. Being able to explore potential combinations by means of computer modelling or simulation could bring us to a much better starting point for initiating clinical trials.
We're also exploring the application of generative AI in choosing drug targets. It can take several years for a targeted drug to be discovered and developed –– locating a target, building out the molecule and executing the trials –– all before it reaches the clinic. Generative AI could help us to home in on the targets and the molecules, enabling faster and potentially more effective predictive modelling. Combining R and generative AI in this space can automate analyses and generate biomarker research more efficiently, so that we can get treatments into the clinic and to patients sooner.
Determining the patients who have the greatest benefit from treatment (responders) or predicting those patients that will not respond or become resistant to treatment is an important factor for successful oncology drug development. With the addition of generative AI on top of multimodal datasets (e.g., clinical, biomarker, ‘omics data), we are more likely to be able to detect such patterns of response or resistance.
Great promise with some limitations
There are limitations to what these new tools can do, and human intellect and brain power will always be required to properly use them. There are also challenges inherent with embracing the change that generative AI is bringing, in particular, and we’ll need to work through several key areas:
- Adapting the workforce and hiring talent dedicated to developing generative AI models and properly leveraging their output
- Finding ways to preserve ownership of data and navigate data privacy
- Considering the amount of energy needed to generate new models from a sustainability perspective
- Communicating specifically and objectively to unlock generative AI’s potential and to avoid so-called hallucinations –– the phenomenon where AI algorithms and deep learning neural networks produce outputs that are not real, and do not match any data an algorithm has been trained on or any other identifiable pattern
Along with excitement and opportunity, generative AI brings new ethical considerations, and at AstraZeneca we are optimistic about maximising the benefits of AI while embodying our company values and following principles for using data and AI ethically.
The future of generative AI at AstraZeneca
Delivering on our commitment to push the boundaries of science and advance innovation, we’ve been early adopters of AI for years. This has helped us speed up drug discovery, identify disease biomarkers, facilitate diagnostics and more. Generative AI, particularly when combined with tools like R, now has the potential to bring us even further, revolutionising how we harness and utilise data, improving the design of clinical trials and guiding our R&D processes. It has great promise to unlock opportunities that allow us to deliver the best possible outcomes for patients, and we are excited about the chapter ahead.
