"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis AI Discovered Antibiotics: How Small Data & Small GNNs Led to Big Results, w/ MIT Prof. Jim Collins
45 snips
Oct 14, 2025 Jim Collins, Termeer Professor at MIT, leads a groundbreaking AI project in antibiotic discovery. He highlights how small datasets and graph neural networks have identified new antibiotics effective against resistant strains. Collins explains the process of screening vast chemical spaces using AI and discusses the staggering impact of antibiotic resistance, calling for public and philanthropic support. He envisions a future where AI extends beyond antibiotics, tackling other therapeutic areas while addressing safety concerns related to drug development.
AI Snips
Chapters
Books
Transcript
Episode notes
Small Data, Big Impact
- Jim Collins' team used small datasets and modest compute to discover new antibiotics.
- Small GNNs trained on ~2.5k compounds can screen tens of millions of molecules quickly.
Bootstrapped Discovery Of Halicin
- Collins bootstrapped an initial screen from a 2,500 compound library built from FDA drugs and natural compounds.
- That pipeline led to the discovery of Halicin as a novel antibiotic hit.
Binarize To Amplify Signal
- They binarized growth inhibition to a 0/1 label to strengthen the learning signal.
- This simplification let models learn structural features tied to antibacterial activity from limited data.
