Top-down design of protein architectures with reinforcement learning

As a result of evolutionary selection, the subunits of naturally occurring protein assemblies often fit together with substantial shape complementarity to generate architectures optimal for function in a manner not achievable by current design approaches. We describe a top-down reinforcement learning- based design approach that solves this problem using Monte Carlo tree search to sample protein conformers in the context of an overall architecture and specified functional constraints. Cryo-electron microscopy structures of the designed disk-shaped nanopores and ultracompact icosahedra are very close to the computational models. The icosohedra enable very-high-density display of immunogens and signaling molecules, which potentiates vaccine response and angiogenesis induction. Our approach enables the top-down design of complex protein nanomaterials with desired system properties and demonstrates the power of reinforcement learning in protein design.

Automated summary

Due to evolutionary selection, naturally occurring protein assemblies have subunits that fit together with substantial shape complementarity to create optimal architectures for function. We present a top-down reinforcement learning-based design approach that utilizes Monte Carlo tree search to sample protein conformers while considering overall architecture and specific functional constraints. Cryo-electron microscopy structures of designed disk-shaped nanopores and ultracompact icosahedra closely resemble computational models. The icosahedra facilitate high-density display of immunogens and signaling molecules, enhancing vaccine response and angiogenesis induction. Our approach allows for top-down design of complex protein nanomaterials with desired properties and exemplifies the power of reinforcement learning in protein design.