A Deep-Learning Neural Network Potential Accelerated First-Principles Study on the Structural Changes Modulated by Methylation and Solvation in 27 Protonated Tripeptides.

Exploring low-energy conformers of tripeptides with different side chains using first-principles methods is important not only to interpret, validate, and predict experimental infrared (IR) spectra in the gas phase but also to understand how the relative stability of different conformations can be modulated by the interplay of basic molecular interactions. In this work, we identified low-energy conformers of 27 protonated tripeptides at M06-2X/6-311+G(d,p) by employing a deep-learning-based neural network potential (DL-NNP) to speed up the structure search. Our methodology also demonstrates a seamless transition from gas phase to implicit-solvent models using the polarizable continuum model (PCM), achieving a mean absolute error (MAE) of energies less than 1.1 and 2.1 kJ/mol, respectively. We found that the number of distinct minima below 25 kJ/mol for a given tripeptide ranges from 10 to 59 in the gas phase and 60-361 in PCM-water. Analysis of the structures of these low-energy minima reveals how methylation modulates molecular interactions through both electronic and steric effects. Finally, the low-energy conformers of methylated tripeptides identified in this study provide valuable insights to compare with available experimental data and to stimulate future experimental and theoretical investigations.