Abstract
<jats:p>Residue-level MM/PBSA and MM/GBSA decompositions are often treated as stable residue rankings, but our results show that apparent disagreement can arise from protocol changes and from reordering many small contributions. We tested this in the PC1/p17 peptide–protein complex and diclofenac–COX2 using matched snapshots, paired residue-energy differences, moving-block bootstrap intervals, rank correlations, and top-k resampling. In PC1/p17, two closely related generalized Born protocols gave identical top-20 sets. Generalized Born versus Poisson–Boltzmann comparisons gave whole-table Spearman correlations of 0.249–0.319. After excluding very small contributions at a 0.25 kcal mol-1 cutoff, the correlations rose to 0.894–0.940. For the GB/PB contrasts, accounting for serial correlation widened median uncertainty intervals by 31–33% but did not change threshold membership. Resampling identified a 17-residue top-20 core. Diclofenac–COX2 showed close agreement between related GB protocols but a different GB/PB workflow-level ordering. DecompStat v0.1.3 was used for the analysis.</jats:p>