TABLE 4.

Match values and bootstrapping standard deviations for the prediction of the sign of the experimental activity differences by single-point mutations of the peptide core amino acids for five state-of-the-art methodologies and the SM-HB-BE (i.e., scoring matrix from hydrogen bonds using half of the conformations with best energies). In addition, we include the computational costs, in days, for running the methods with the 56 pairs of mutated peptides. The strategies are the sequence motif matrix, the machine learning tool NetMHCIIpan, the MD/scoring and backrub/scoring approaches, and the MM-PBSA calculations (see Methods).

Complementary strategy	Matched predictions	Computational cost (days)
Sequence matrix	0.393 ± 0.067	0.05
NetMHCIIpan	0.536 ± 0.079	0.1
Backrub/scoring	0.536 ± 0.067	2
MD/scoring	0.571 ± 0.071	15
MM-PBSA	0.518 ± 0.062	15
SM-HB-BE scoring matrix	0.589 ± 0.065	0.05