Network pharmacology

VOLUME 25 NUMBER 10 OCTOBER 2007 NATURE BIOTECHNOLOGY the wild-type complexes that, on burial from solvent, pay an electrostatic desolvation penalty that is uncompensated by the interactions that they make in the complex. Such a distinction is something to which the physics-based electrostatic model deployed in the calculations is well suited. Thus, most successful predictions involved the substitution of polar with apolar residues, which will be less penalized for desolvation on complex formation. Where new polar interactions were also introduced, for instance, by replacing a wild-type threonine with an asparate, these occurred on the periphery of the complex where detailed complementarity is less important. A pragmatic lesson may be that one can often improve affinity more by reducing desolvation penalties than by optimizing interaction energies. By the same standard, the lack of correlation between the full energy function and the experimental interaction energies makes clear that this work is not a general solution to the problem of calculating ligand-protein binding affinities. As the authors acknowledge, the geometries of the complexes were refined by calculation using the full energy function, which was then discarded in favor of the electrostatic terms alone. This cannot be right, and may be expected to fail in many cases. Indeed, whereas single-site substitutions were well predicted in this study, calculations on multi-site substitutions correlated poorly with experiment, as did predictions on the D1.3 antibody–lysozyme complex. Notwithstanding these caveats, the results reported by Lippow et al. are encouraging. Their predictions derive from a general physics-based model that should be widely applicable to the electrostatic component of binding affinity, and the calculations were rapid enough to guide an experimental program of affinity maturation. The problems encountered here may be addressable by greater sampling of the states and configurations open to proteins and their ligands; methods to do so have been developed, although they remain slow enough to prohibit application to a large number of possible ligands. The calculations in Lippow et al. balance physical rigor with practicality and may be useful for optimizing other protein-protein complexes.