Cette session rassemble des expertes et experts qui s’efforcent d’établir les critères de référence pour mesurer les progrès, relever les défis et bâtir une communauté en matière de découverte computationnelle de médicaments.

«Rethinking Rigor: Towards Community Standards for the Validation of AI Models in Biomedicine»

Gustavo Stolovitzky - Dir. Biomedical Data Sci. Hub at NYU Langone Health & Founding Chair, DREAM Challenges

The explosive growth of data-driven artificial intelligence in biomedicine is reshaping the scientific landscape, powering innovative healthcare solutions and transformative insights in life sciences. AI is proving to be a new paradigm for predictive modeling, which in turn demands a paradigm shift in our approach to model evaluation. Traditional algorithmic performance metrics alone are insufficient for comprehensive validation. Concurrent with the rise of AI modeling opportunities has been a proliferation of community-led benchmarking initiatives, fragmenting efforts and limiting their collective synergy and impact. This talk proposes fostering a coalition of Critical Assessment efforts. Such a coalition can enhance collaboration, establish rigorous validation standards, and build sustained trust in evolving AI methodologies, ensuring their robust and reliable integration into biomedical research.

«Popularizing best practices through the Polaris benchmarking platform»
Cas Wognum - Sen. Scientist at Valance Labs

Machine learning (ML) is driving innovations in drug discovery, but we need to be mindful of the circumstances that set this application apart. The absence of standardized, domain-appropriate datasets, guidelines and tools for the evaluation and comparison of methods has led to a growing gap between perceived progress and real-world impact, delaying the adoption of ML in drug discovery. To bridge this gap, we formed Polaris: An open-science, cross-industry and interdisciplinary collaboration to develop benchmarking protocols tailored to drug discovery. With Polaris, we go beyond publishing papers and develop a multimodal benchmarking platform for machine learning in drug discovery to simplify the adoption of these protocols. This talk will share some of the guiding design principles we followed when building Polaris and discuss the role of open-source software in popularizing appropriate benchmarking protocols.