Written by:
Executive Director, Data Science for Early Oncology, AstraZeneca
Executive Director, Oncology Data Science Platforms, AstraZeneca
Executive Director, Real World Evidence, AstraZeneca
Artificial intelligence (AI) is evolving into a powerful tool for helping scientists develop new, innovative treatments for cancer.1 While progress is being made, to fully realise the potential of AI we need to maximise the use of our clinical and multi-omics data. To enable this, our Data Science teams are optimising our data and AI foundations.
Transforming oncology in the AI generation
The biotech and pharmaceutical industry is in a transformative era driven by a convergence of science, data and tech.1 The growing accessibility of computational tools and next-generation sequencing is fostering a thriving clinical and biological data ecosystem.1 With the help of AI, researchers can tap into this ecosystem to speed up drug discovery, identify disease biomarkers and facilitate diagnostics.1
One of the most significant applications of big data and AI is in oncology.2 The promise of precision medicine is now a reality, with therapies increasingly targeted to different patient populations depending on the underlying genetic makeup of their disease.2 However, cancer is a complex biological problem, particularly tumours resistant to therapy or evolving to acquire resistance over time.3
There are vital clues for designing innovative cancer treatments hidden within the industry’s ever-growing biological and clinical data collections. To maximise the value of these data, individual datasets must be united and organised, allowing data science and AI to drive new ideas for cancer drug development.
Building up our oncology data foundations
We hold a vast trove of oncology data from more than 100,000 consenting patients, including clinical, imaging, and multi-omics data. Established last year, the company’s Oncology Data Science team feeds these data into a system that uses AI and other statistical tools to generate novel hypotheses in oncology drug development.
To achieve this transformation, the team is adapting our complex datasets to make them findable, accessible, interoperable, and reusable according to a set of principles collectively known as FAIR.4 This allows data collected from specific trials and projects to be accessible across the company’s drug development teams in full accordance with data protection laws worldwide.
In addition to data from our clinical trials, we are working with external companies such as Tempus to leverage real-world data, representing patients from around the world. The strategic partnership provides crucial evidence about patient outcomes in the health system without revealing the identity of the patients in the datasets.
AI models and validation: a virtuous circle
With the data organised according to FAIR principles, the Oncology Data Science team is leveraging the latest ML techniques to construct models that guide drug development efforts and the efficient design of clinical trials. For example, the team is using knowledge graphs to integrate millions of data points to produce novel target and disease insights, and applying transformer models to identify drug response biomarkers.
A key goal for the Oncology Data Science organisation is to decode cancer and provide actionable results to scientists and physicians at the right time so that key decisions can be made with the right data set, from designing clinical trials to selecting drug targets. Another is that bench scientists and clinicians can validate AI-based biological predictions through lab studies and trials, generating data that can be fed back into the model. This creates a virtuous circle of AI-guided hypothesis generation and validation.
This emerging field, often termed computational oncology will stimulate innovation in our portfolio, uncovering new insights and evidence in the field of cancer. Our data scientists are working hard to unpick disease mechanisms, interrogate new cellular pathways in oncology, and deliver novel targets for our pipeline which have the potential to be addressed with our breadth of treatment platforms, from antibody-drug conjugates through to T-cell engagers.
The future of data science in oncology
Data science and AI are becoming increasingly important in drug discovery and development. In addition to generating new therapeutics, AI speeds up the work of existing drug discovery and clinical teams, helping them to make informed and highly accurate research decisions.1
To go beyond this point over the coming decade, we must ensure we have solid data foundations and tight circles of model validation to leverage our data resources. We’ve already seen an AI-guided approach bear fruit with our first disease models based on knowledge graphs focussed on understanding drug resistance.
We have high hopes for the progress we will make as this field continues to accelerate into the future.
