Datasets:

ArXiv:
License:
timewarp / 4AA-large /README.md
andrewfoong's picture
Upload folder using huggingface_hub
4030900 verified

4AA-large "Four Amino Acid" data set, tetrapeptides

The difference to the 4AA data set is the much longer simulation time. This folder contains a data set of all-atom molecular dynamics trajectories for 2333 tetrapeptides, i.e. small proteins composed of four amino acids. The data set is split into 1500 tetra-peptides in the train set, 400 in validation, and 433 in test. Each peptide in the train set is simulated for 50ns using classical molecular dynamics and the water is simulated using an implicit water model. Each other peptide is simulated for 500ns.

For each trajectory two files are available:

  • protein-state0.pdb: contains the topology and initial 3D XYZ coordinates.
  • protein-arrays.npz: contains trajectory information.

NPZ Information

Each NPZ file contains detailed information for a subset of simulation steps. There are T such frames and each NPZ file contains the following arrays:

  • 'step': (T,) array, Md step number.
  • 'energies': (T,2) array, each row containing [potential, kinetic] energies in kJ/mol.
  • 'positions': (T,num_atoms,3) array, positions in nm.
  • 'velocities': (T,num_atoms,3) array, velocities in nm/ps.
  • 'forces': (T,num_atoms,3) array, forces in kJ/(mol nm).

Dataset construction

The dataset was constructed in the following way:

  1. For all included PDB files, perform a molecular dynamics simulation:

    a.) Use OpenMM with the AMBER14 force field and implicit water (preset=amber14-implicit).`

    b.) Perform an energy minimization (relaxation) from the initial PDB configuration. c.) Use a Langevin integrator at temperature T=310K, friction=0.3/ps, timestep=0.5fs for 2e6 steps to equilibriate ("burn-in phase"). d.) Use a Langevin integrator at temperature T=310K, friction=0.3/ps, timestep=0.5fs for 1e8/1e9 steps to sample a trajectory ("sample phase").

  2. Save trajectory information every 10,000 steps (5ps) to an arrays.npz file.

Credit and Authors

This dataset was created in September 2022 as part of the Molecular Simulation initiative.