Principles for computational design of binding antibodies

Significance Antibodies are the most versatile class of binding molecule known, and have numerous applications in biomedicine. Computational design of antibodies, however, poses unusual difficulties relative to previously designed proteins, as antibodies comprise multiple nonideal features, such as long and unstructured loops and buried charges and polar interaction networks. We developed an algorithm that uses information on backbone conformations and sequence-conservation patterns observed in natural antibodies to design new antibody binders. Designed antibodies were very different in sequence from natural ones, but had similarly desirable properties of affinity and stability, and molecular structures showed high accuracy relative to the design models. The design principles we implemented can be used to design other functional folds, including many enzyme classes.