Design of Protein Sequences with Precisely Tuned Kinetic Properties

Recent advances in computational biology have enabled solutions to the inverse folding problem - finding an amino acid sequence that folds into a target structure. An open question concerns the design of proteins that in addition to having the correct target structure also have precisely tuned kinetic properties, such as folding and unfolding rates. To address this problem, we formulate the inverse folding problem as a quest for a sequence with a target free energy landscape. To propose a procedure to address this problem, here we describe the Inverse Folding Molecular Dynamics (IF-MD) method, which combines inverse folding with enhanced sampling molecular dynamics and Bayesian optimization. IF-MD leverages ensemble averages from molecular dynamics simulations, reweighted according to a Bayesian framework, to guide the design of sequences exhibiting specific kinetic properties. We demonstrate the methodology by optisising the binding kinetics of H11, a nanobody against the SARS-CoV-2 spike receptor-binding domain (RBD), thus identifying nanobody variants with slower unbinding kinetics than H11. Mechanistic analysis reveals that this kinetic property arises from a shift towards configurations closer to the bound state and increased free energy barriers for dissociation. These findings highlight the power of IF-MD for efficiently navigating the vast sequence space to design proteins with a tailored free energy landscape.