The covalent nature of interactions within various hydrogen bonded molecular aggregates has been characterized by the two entirely different computational methods: Bader analysis of the electron density and variation-perturbation partitioning of the intermolecular interaction energy. Analysis of 34 complexes representing different types of hydrogen bonds indicates that the proton-acceptor distance approximately 1.8 A and the ratio of delocalization and electrostatic terms approximately 0.45 constitutes approximately a borderline between covalent and noncovalent hydrogen bonds. The latter ratio could be used to characterize quantitatively the degree of the covalent nature of transition state interactions with active site residues, a quantity essential for an enzyme catalytic activity.

Interaction energies of phenylalanine ammonia-lyase (PAL) active site residues with a series of PAL inhibitors have been partitioned into electrostatic, exchange, delocalization, and correlation components and compared with analogous results obtained previously for leucine aminopeptidase (LAP). In the latter metalloenzyme, either of the two charged residues controls entirely relative inhibitor binding energies, while at least four residues are required to determine ligand relative stabilization in neutral PAL. Significant correlation with experimental inhibitory activity was found between the stabilization energy at gradually decreasing levels of theory (MP2, SCF) down to the first-order Heitler-London term. Contrary to the LAP case, where the electrostatic term was sufficient to reproduce experimentally observed trends, in the case of PAL, exchange repulsion effects also have to be considered. Computational protocol presented herein constitutes a promising way to incorporate the first principle calculation's accuracy into the process of rational binding affinity prediction, revealing the physical nature of the interactions, where successive approximations can be introduced in a systematic and justifiable manner.