BPS2026 – Computational design of protein allostery

The ability to design a novel sequence to adopt a specific fold has improved significantly over the last decade. This is largely due to the advent of AI tools, such as RFdiffusion, ProteinMPNN, and AlphaFold. This has opened up protein design to numerous applications, from enzyme design to protein therapeutics. However, protein design still struggles in the design of protein dynamics. This flaw is particularly apparent as it is becoming ever clearer that proteins derive much of their function from their dynamics. Protein function arises through precise control of a protein's energy landscape. Kinases, G-proteins, and other signaling molecules react to stimuli by propagating responses through conserved dynamic networks while enzymes use dynamics and conformational selection to drive turnover. Here, we design de novo proteins which exhibit dynamics through allostery. Through inspiration from natural dynamic proteins, we repurpose existing deep-learning methods to computationally design multiconformational proteins and verify these conformations experimentally. The ability to design specific protein energy landscapes and dynamics would greatly increase the scope of protein design and would allow for the design of protein biosensors and tunable protein functions.