The AI in Business Podcast Building AI-Ready Cultures in Life Sciences R&D - with Xiong Liu of Novartis
Oct 21, 2025
Xiong Liu, Director of Data Science and AI at Novartis, discusses how generative AI is revolutionizing life sciences R&D. He highlights the shift from traditional machine learning to foundation models, which can significantly enhance small-data tasks. Xiong explains the importance of domain-specific strategies and the role of data quality in accelerating research. He emphasizes organizational alignment among scientists and AI teams to ensure successful AI adoption, providing insights on overcoming challenges like data scarcity and validation risks.
AI Snips
Chapters
Transcript
Episode notes
Foundation Models Extend Small Datasets
- Foundation models let teams leverage large, diverse domain data so small task datasets get useful background knowledge.
- This reduces the need for vast labeled data per task and enables many downstream applications in R&D.
Domain Priors From Public Molecular Data
- Domain-specific foundation models provide biological priors like gene interactions and cell-type patterns from public molecular data.
- Fine-tuning those models with proprietary data increases relevance and prediction accuracy for specific indications.
Constrain Generative Models For Molecules
- Represent molecules as sequences or graphs and pretrain transformer models on public chemical and biological databases.
- Constrain generation with toxicity and solubility objectives to produce viable candidate molecules.