Takeways - AMLDEPFL2024
Our conference track on "Machine Learning for Rare Disease Applications" brought together experts from various domains to explore the challenges and advancements in applying data science, particularly in clinical settings dealing with rare diseases.
Throughout the discussions, several key insights emerged:
- Impressive Technological Applications
Speakers showcased the breadth of technological advancements, from leveraging computer vision for microbiology to employing sophisticated statistical and generative AI methods in drug development and pediatric oncology. These examples underscored the versatility and potential impact of machine learning (ML) across various domains within healthcare. - Trade-off between Regulations and Research
We highlighted the intricate balance between evolving regulations and the pursuit of innovative research. New regulations can sometimes impede certain types of research endeavors, necessitating adaptability and collaboration between regulatory bodies and researchers. - Diverse Perspectives on Generalisation
There exists a spectrum of opinions regarding the importance of model generalisation. While some individuals, particularly those in computer science, prioritise robustness and generalisation across contexts, others, particularly in clinical settings, may prioritise effectiveness within specific contexts. Understanding and reconciling these differing viewpoints is crucial for effectively applying ML techniques in clinical practice. - Challenges Specific to Rare Disease Applications
The unique nature of rare diseases presents specific challenges, including limited data availability, the inherent need for model generalizability across diverse patient populations, and the requirement for synthetic controls to support randomized controlled trials (RCTs). - Diverse Approaches to Address Challenges
Throughout the session, various approaches were discussed to tackle the challenges in rare disease applications. These approaches ranged from purely data-driven methodologies, as advocated by Anshul, to hybrid approaches combining data-driven techniques with mechanistic modeling, as demonstrated by Sebastien and Sobhan. The relevance of these approaches was evaluated from the perspectives of clinical experts (Adrian), industry (Li), and specific application contexts (Daria).
We extend our sincere gratitude to all the contributors to this enlightening session: Adrian Egli (University of Zurich 🇨🇭), Anshul Thakur (University of Oxford 🇬🇧 ), Ana Daria Laslo (Department of Health Sciences and Technology (D-HEST), ETH Zürich 🇨🇭), Li Wang (AbbVie 🇺🇸 ), Sobhan Moazemi Goodarzi (Frauenhofer SCAI 🇩🇪 ) and Sebastien Benzekry (Inria 🇫🇷 )
Proud to have brought this group together and already excited for the follow-ups.
Great job team, Catherine Jutzeler, Sarah Brüningk, Samuel Håkansson, Ana Daria Laslo, Aspasia Vozi, Lakmal Meegahapola, Louis Philipp Lukas.
Link to the conference external page AMDLEPFL2024.
Visit us in external page LinkedIn BMDS for more news.