393 AB-Bind : Antibody binding mutational database forcomputational affinity predictions

Potent antibodies are needed for applications in research and medicine, and computational tools can potentially accelerate antibody engineering. We have developed the Antibody-Bind (AB-Bind) database, which compiles a diverse set of antibody binding data and accompanying structures to enable method evaluation for modeling antibody interactions. Using the AB-Bind data set, we studied the performance of protein scoring potentials for predicting changes in binding free energies upon mutation. Numerical correlations between computed and observed DDG values were low for all tested methods, but the potentials do have utility for classifying variants as improved vs. weakened binders.