SwitchCraft: A Programmatic Framework for Designing State-Switching Proteins

Multistate mechanisms underlie many of the complex functions observed in natural proteins. The ability to rationally design multistate proteins would have transformative implications for many areas of biotechnology, yet lies beyond the capabilities of existing deep learning frameworks for protein design. To address this gap, we introduce SwitchCraft, a versatile and programmatic framework for designing state-switching proteins based on backpropagation through compositional design constraints parameterized by structure prediction models. In silico evaluations demonstrate success on a wide range of state-switching functional primitives, from allosteric regulation of motifs to discrimination of bound ligand identities. Using these primitives, we demonstrate an in silico strategy for de novo design of fluorescent biosensors to arbitrary small molecule analytes. These results position SwitchCraft at the inception of a powerful paradigm for higher-order functional protein design. Code is available at https://github.com/bjing2016/switchcraft.

Proteins are tiny molecular "machines" that carry out a wide array of tasks within our cells to sustain life. Generative models have given researchers powerful tools for designing new proteins beyond the proteins that exist in nature. However, most existing models can only design proteins with simple functions, like sticking to another protein, or copy functions from proteins that already exist. We develop a deep learning method to design proteins that change shape as part of their function, which underlies the operation of many of natural proteins. We found that our method is versatile and can replicate many of the shape-shifting behaviors that underlie natural protein functions, controlled in a way that the user specifies. We think our method could eventually help researchers develop new biotechnologies and therapeutics by providing greater control over molecular behavior.