The Timewarp dataset, released by Microsoft Research in 2023, is designed to accelerate molecular dynamics simulations. The associated research paper is "Timewarp: Transferable Acceleration of Molecular Dynamics by Learning Time-Coarsened Dynamics". It aims to use deep learning techniques to predict the three-dimensional conformational changes of small peptides, thereby accelerating the molecular dynamics simulation process.

The dataset contains extensive molecular dynamics trajectory data for small peptides (2-4 amino acids), simulated using an implicit water field. It covers various small molecular systems, including dipeptides and tetrapeptides. The dataset includes protein state files and trajectory array files, supporting the training and validation of machine learning models in the field of molecular dynamics. It provides an important data foundation for studying protein folding and kinetic behavior.

Dataset Composition

• 2AA-1-big: Contains all-atom molecular dynamics trajectories for 380 dipeptides, generated using classical molecular dynamics and an implicit water model, with trajectories saved every 10,000 steps.
• 2AA-1-complete: Contains complete trajectory data for all 400 dipeptides, supplementing the 20 dipeptides missing from 2AA-1-big. It also employs classical molecular dynamics and an implicit water model.
• 4AA-huge: Contains molecular dynamics trajectories for tetrapeptides, primarily used for validation and testing, with a simulation duration of 1 microsecond.
• 4AA-large: Contains 2,333 tetrapeptides, divided into a training set (1,500), a validation set (400), and a test set (433). The training set has a simulation duration of 50 nanoseconds, while the validation and test sets have a duration of 500 nanoseconds.
• AD-3: Contains two long trajectories simulated for alanine dipeptide, a molecule consisting of 22 atoms.