ペプチド結合自由エネルギーの正確なインシリコ予測のための接触並列カスケード選択分子動力学（cPaCS-MD）

Although peptide drugs make up a growing proportion of the pharmaceutical market, computational prediction of peptide-protein binding affinity remains challenging, which limits the use of computational methods to assist the optimization of bioactive peptides. We have developed a novel method for modeling peptide unbinding, designated contact parallel cascade selection molecular dynamics (cPaCS-MD), which, when combined with Markov state models, accurately predicts peptide binding affinities. We have applied cPaCS-MD to a diverse set of 12 protein-peptide complexes and found that it predicts experimental peptide binding free energies with a strong correlation (R² = 0.84) and high accuracy (mean absolute error of 2.7 kJ/mol and root-mean-square error of 3.4 kJ/mol). We compared cPaCS-MD to the widely used umbrella sampling (US) method and found that PaCS-MD is more computationally efficient and has much higher accuracy. This work is the first benchmarking study of MD methods for predicting peptide binding affinities that uses a diverse set of peptide ligands. We expect that cPaCS-MD will become a widely used and versatile tool for the study of peptide-protein unbinding and will assist the development of peptides for diverse applications.