Dataset Peptoid Molecular Structures: Configurations, Machine-Learned Interatomic Potentials, and Molecular Dynamics Simulations

PEPPER data and models

This release contains the PEPPER MLIP training datasets, trained machine-learning interatomic potentials, and associated molecular-simulation/evaluation workflows for peptoid conformational energetics.

• training-data/ contains the MIN.extxyz, ALL.extxyz, and HCAPPED.extxyz training datasets.
• SRC-MLIPs/ contains the custom-trained DeepMD and MACE models, together with selected pretrained MLIPs used for comparison.
• ML-MD/ contains the per-residue LAMMPS workflows used to evaluate MLIP energies and forces for the peptoid conformations.
• README.md provides documentation for the dataset organization, models, and associated workflows.

MIN.extxyz

Minima-only PEPPER training dataset containing 3,366 representative peptoid conformations selected from low-energy cis and trans conformational basins while retaining diversity in backbone configurations and side-chain rotamers.

ALL.extxyz

Transition-augmented PEPPER training dataset containing 6,153 configurations: the 3,366 low-energy structures in MIN.extxyz plus 2,787 higher-energy structures sampling the cis/trans transition region.

HCAPPED.extxyz

Hydrogen-capped, sidechain-focused PEPPER training dataset constructed to examine the importance of the complete peptoid backbone environment. Truncated backbone connections are terminated with hydrogen atoms, providing a simplified molecular representation for comparison with the full peptoid structures. Full disarcosine (R01) is included, all other side-chains have been capped by hydrogen to compare against literature approaches.

Bradley S. Harris, Mohammadhasan Dinpajooh, Marcel D. Baer, Pacific Northwest National Laboratory, Richland, Washington, USA