Zero-shot design of drug-binding proteins via neural iterative selection-expansion.

Neural iterative selection-expansion (NISE) enables zero-shot design of small-molecule binding proteins with nanomolar affinities, achieving success rates up to 100%.

• Why it matters: Designing proteins that bind small molecules is complex due to the need to optimize sequences, structures, and ligand conformations simultaneously, a challenge that current deep-learning methods have struggled to overcome.
• What they did: The authors combined a ligand-aware neural network (LASErMPNN) with a structure predictor in an iterative loop, training and pairing these models to optimize protein-ligand compatibility for different drug targets, including exatecan and apixaban.
• The result: This approach produced high-affinity binders, surpassing previous methods by up to 70,000-fold in affinity, and enabled affinity improvements through neural suggestions, demonstrating broad potential for drug delivery, sensing, and catalysis applications.