The deformation energy gap in computational drug design: why interaction energy alone cannot rank drug candidates and...

DFT calculations are increasingly combined with molecular docking to rank drug candidates, yet most studies report the interaction energy (ΔE_int), computed at the complex geometry, as a surrogate for binding affinity. This quantity omits the deformation energy (ΔE_def): the thermodynamic penalty of distorting both partners from their free-state geometries into their bound conformations. Because ΔE_def is always positive (typically 2-20 kcal mol^-1) and molecule-dependent, its omission systematically overestimates binding strength and can reverse predicted rank-orderings. We present the energetic decomposition ΔEbind = ΔEint + ΔEdef, demonstrate using published crystallographic strain data from over 3,000 protein-ligand complexes that deformation energies do not cancel between structurally distinct ligands, and propose a minimal five-step correction protocol applicable to any DFT-based drug design study. The protocol requires only two additional geometry optimizations beyond the standard workflow, adding only modest additional computational cost. This work does not introduce new computational data; it highlights an energetic inconsistency in common computational practice and provides a straightforward correction to enable more consistent electronic binding energy evaluation and improved candidate comparison.  METHODS: The analysis is based on the supramolecular energy decomposition framework and the activation strain model (ASM), in which binding energy is partitioned into interaction and deformation (strain) components using standard variational principles. No new DFT calculations are reported. The argument draws on published conformational strain datasets obtained at various DFT levels and molecular mechanics force fields from crystallographic analyses of the PDBBind database. The proposed correction protocol is general and can be applied with any DFT functional, basis set, and quantum chemistry software package (e.g., Gaussian, ORCA, or equivalent).