How pharma can incorporate AI into research
The use of artificial intelligence (AI) in the biopharma industry has the potential to revolutionise drug discovery and development, as well as improve patient care. By analysing large amounts of data, AI algorithms can identify patterns and relationships that may not be apparent to humans, leading to new insights and therapies.
One way that biopharma companies are incorporating AI into their research is through the use of federated learning. Federated learning allows multiple organisations to work together on a machine learning model without sharing their data, by training models locally and then aggregating the results. This approach was demonstrated to be effective in the EXAM study, published in Nature Medicine in 2021, which showed that federated learning could be used to identify patients at heightened risk of mortality and serious morbidity due to COVID-19 (disclosure: this author was the first author of the publication).
Some of these provider-led collaborations, with EXAM being a prominent example, were ad hoc in nature, forming quickly in response to the crisis. As a result, some struggled to achieve long-term sustainability and scale, as they did not have the necessary infrastructure or support in place to continue operating once the immediate crisis had passed. However, the seed of using federated learning to collaborate on machine learning (ML) is now planted, and is taking its first steps on the path of becoming the new standard for institutional and industry/provider collaborations.
Overcoming challenges through collaboration & federated learning
The biopharma industry is now emerging as an early adopter of this technology, looking to leverage it for overcoming the challenges of accessing real-world data, which is often fragmented and siloed, and even in order to work together with other biopharma companies, and using their data resources in order to gain a more comprehensive understanding of diseases and treatments for mutually beneficial purposes. An early example of that is the MELLODDY consortium, in which biopharma companies collaborate on specific research projects, with some European healthcare organisations involved as well. Since then, biopharma companies have expanded their use of federated learning to an increasingly sophisticated and diverse set of uses, from covering drug discovery by studying molecular interactions, to identifying patients for clinical trials, and onto early detection of disease and drug repurposing.
Regardless of these new opportunities, using federated learning as a way to provide more data to the scientists’ ‘workbench’, comes with challenges. Adapting biopharma research workflows to use federated learning requires solving compatibility issues with different data formats, ensuring data quality, data security concerns, network infrastructure, legal and regulatory compliance, and building trust among organisations. Expanding on some of these:
- Data compatibility: Different biopharma research organisations may use different data formats or data structures, making it difficult to combine and use the data for federated learning. This may require significant effort to pre-process the data and ensure compatibility.
- Data quality: Ensuring that the data used for federated learning is of high quality and accurate can be a significant challenge. This may require significant effort to validate and clean the data before it can be used for model training.
- Data security: Federated learning relies on the ability to share models across different organisations, which can raise concerns about data security and privacy. Ensuring that the data is protected and that only authorised parties have access to it can be a significant challenge.
- Network infrastructure: Federated learning requires a robust network infrastructure to enable the sharing of models and data across different organisations. This can be a significant challenge for organisations that do not have the necessary infrastructure in place.
- Legal and regulatory compliance: Federated learning may raise legal and regulatory compliance issues, such as data privacy and data protection laws. Organisations will need to ensure that they are compliant with these laws and regulations before implementing federated learning.
- Collaboration and trust: Collaboration is the key aspect of federated learning, however, building trust among different organisations to share models and data can be a challenge and requires experience and specialised frameworks, to ensure all sides feel at ease.
- Data science skills: Lastly, while scientists in biopharma companies have taken a big leap forward in terms of data science and different technical proficiencies, the need to use distributed compute systems, such as the ones that support federated learning, is yet another requirement and can slow down adoption due to lack of skilled workforce.
Overall, while federated learning has the potential to greatly benefit biopharma research, it is not a simple process to implement and will require a significant effort to address these challenges.
Nonetheless, these challenges are driving the emergence of federated learning as a new competence, in which biopharma is investing in, and a new category of companies that are dedicated to driving the adoption of distributed compute systems.
About the author
Ittai Dayan is the co-founder and CEO of Rhino Health. His background is in developing artificial intelligence and diagnostics, as well as clinical medicine and research. He is a former core member of BCG’s healthcare practice and hospital executive and is currently focused on contributing to the development of safe, equitable, and impactful AI in healthcare and the life sciences industry. Dayan served in the IDF special forces and is a long-distance runner.