NEJM AI Grand Rounds Bridging AI and Biology to Tackle Medicine’s Hardest Problems with Dr. Marinka Zitnik
12 snips
Jan 21, 2026 Dr. Marinka Zitnik, an associate professor of biomedical informatics at Harvard, is a pioneer in using AI for drug discovery. In their conversation, she highlights the challenge of treating rare diseases and shares how AI-driven drug repurposing can accelerate solutions. Marinka emphasizes the need for collaboration between machine learning and biology to tackle real patient problems. She also defines AI agents as systems capable of learning and taking action, which she believes is vital for advancing scientific discovery.
AI Snips
Chapters
Transcript
Episode notes
Early Breakthrough With Amoeba Gene Clustering
- As a high school student Marinka clustered amoeba gene sequences and produced a ranked list for experiments.
- Baylor collaborators validated nine of her top 12 predictions, showing immediate real-world impact.
Foundation Models Speed Rare-Disease Treatments
- Drug repurposing offers a faster, cheaper path to treatments, especially for rare diseases with little commercial incentive.
- TXGNN trains across thousands of phenotypes to predict drug–disease matches and generalizes zero-shot to diseases without approved therapies.
Two Rationales Behind Drug Repurposing
- Repurposing can be drug-centric (off-target binding) or disease-centric (phenotypic/molecular similarity).
- Machine learning can learn similarity metrics to transfer treatments from well-studied diseases to understudied ones.