Structure of poly-glycine, poly-alanine, and poly-sarcosine: a neural network assisted first-principles study on the influence of methylation and chain length.

In this work, we employed neural network potentials (NNPs) to accelerate the conformational exploration of poly-glycine, poly-alanine, and poly-sarcosine up to hexapeptides at the M06-2X/6-311+G(d,p) level. One methodological advance is that, for penta- and hexapeptides, only single-point DFT calculations (energies and forces) are sufficient to train NNPs to achieve an accuracy of approximately 4 kJ mol-1. With these accurate NNPs, we efficiently identified low-energy (<25 kJ mol-1) minima for all 12 peptides. Based on the structures of these low-energy conformers, we found that: (1) the relative stability of the zwitterionic form increases consistently with peptide size. Among these peptides, the zwitterionic form becomes the global minimum only for hexa-sarcosine. (2) N-methylation shows a strong propensity to stabilize cis-peptide bonds. Tri-, tetra-, and hexasarcosine preferentially adopt cis-peptide configurations. By comparison, Cα-methylation plays a less significant role. Interestingly, tetra- and hexaalanine are more likely to adopt cis conformers compared to penta-alanine, which has recently been studied experimentally via IR-VUV spectroscopy. The DFT-optimized coordinates of all 12 peptides can serve as a useful reference for validating our theoretical results against future experimental IR studies.