Surface-layer proteins (SLPs) are multi-domain proteins that self-assemble into a nanoporous lattice, the so-called S-layer, on the exterior of many bacteria and archaea. The S-layer protects and aids signaling between cells and has been identified as virulence factors in pathogens. Recently, a subset of camelid nanobodies targeting the B. anthracis SLP were found to disarm the pathogen by disassembling the lattice; infections in murine models were cleared within six days, suggesting that nanobodies may be used as therapeutics to prevent pathogenesis while lowering the risk of antibiotic resistance from selective pressure. However, several nanobodies bound to the same site of the SLP did not cause lattice disassembly, suggesting that binding affinity alone cannot account for the functional outcome of all nanobody therapeutics. To design more effective nanobody therapeutics, we first present a machine learning approach to classify the efficacy of eleven high-affinity nanobodies on the basis of their dynamical impact on the SLP using molecular dynamics simulations. We then present a workflow to computationally design nanobodies with improved efficacy (i.e., increased depolymerization rates at decreased nanobody concentrations) using an iterative genetic algorithm approach combined with active learning to limit the number of additional experiments that are required. Finally, we discuss the extension of our approach to other pathogenic bacteria, including C. difficile.
