Programming biological behavior with AI-guided de novo protein design

Abstract

Driven by advances in artificial intelligence, de novo protein design can now generate diverse molecular functions from the ground up, achieving remarkable success in engineering binders against diverse macromolecular targets. Here, we provide a perspective on how de novo design can progress beyond bimolecular interactions toward the bottom-up construction of protein reaction networks. We highlight how programmable, modular, and controllable de novo components can drive complex, emergent biological behaviors. Overcoming current challenges—including designable protein dynamics, multi-step functionalities, and non-equilibrium systems—is essential for this transition. With continued foundational method improvements, the field is poised to pioneer precise protein perturbations for systems-level insight and to engineer emergent cellular behaviors like signal amplification, decision-making, and error correction de novo to fundamentally advance synthetic biology, biotechnology, and medicine.