Alanine dipeptide (coarse grained, with forces)
| Property | Value |
|---|---|
| Code | OpenMM |
| Forcefield | AMBER ff-99SB-ILDN |
| Integrator | Langevin |
| Integrator time step | 2 fs |
| Simulation time | 4x500 ns |
| Frame spacing | 2 ps |
| Temperature | 300 K |
| Volume | (2.3222 nm)^3 periodic box |
| Solvation | 651 TIP3P waters |
| Electrostatics | PME |
| PME real-space cutoff | 0.9 nm |
| PME grid spacing | 0.1 nm |
| Constraints | all bonds between hydrogens and heavy atoms |
Contents
The molecule and system setup is essentially the same as in the other alanine dipeptide dataset. The major difference is that we cropped out the five backbone atoms (ACE-C ALA-N ALA-CA ALA-C NME-N) and recorded the coordinates and forces of them. The dataset contains four independent simulations, each of which consists of 250,000 frames (500 ns sampled with a 2-ps interval). Therefore, for cross-validation it is recommended to separate the data into four folds by indexing the arrays with [250000 * i_fold:250000 * (i_fold + 1)].
Raw data
Shared PDB topology file, coordinates (key coords) and forces (key aaFs) on the five backbone atoms from four independent simulations (four trajectories were concatenated together numpy.ndarray(shape=[1000000, 5, 3], dtype=numpy.float32)). The unit of coordinates is Angstrom (0.1 nm) and for the forces kcal/(mol·Angstrom).
Citations
This dataset was used in the following publication for the first time.
- J. Köhler et al.: Flow-Matching: Efficient Coarse-Graining of Molecular Dynamics without Forces, J. Chem. Theory Comput. 19 (2023), 942.