ITrace

All-atom molecular-dynamics trajectories for 245 experimentally determined class I pMHC–TCR complexes, each simulated in three independent 200 ns replicas — 735 trajectories, 147 µs of aggregate sampling.

Of the 245 source structures, 243 were solved by X-ray diffraction and 2 by cryo-electron microscopy (9rup, 4.11 Å; 9rxm, 3.0 Å).

Trajectory records are identified as <pdb_id>_run<replica>, for example 1ao7_run3. Files must never be paired across records.

An interactive web interface for browsing and downloading these trajectories is available at http://iqb.zju.edu.cn/servers/itrace/.

Layout

full_system/<record>/traj.xtc    solvated system, periodic-boundary corrected
full_system/<record>/md.tpr      matching GROMACS run input
processed/<record>/traj.xtc      protein heavy atoms, backbone-fitted
processed/<record>/topology.pdb  matching topology for the processed trajectory

Both layers are complete for all 735 trajectory records, and are sampled at a nominal 200 ps interval over 0–200 ns of the production run; processed holds exactly 1,001 frames throughout. Exact frame counts, and the atom selection of each layer, are reported for every trajectory in manifest.tsv. full_system retains solvent, ions and hydrogens at the production atom ordering; processed keeps protein heavy atoms only and is the representation from which the released descriptors were computed. Every record's processed and full_system layers were checked against a box-volume provenance fingerprint and agree with their own sibling replicas.

Reading the files

• full_system/<record>/traj.xtc must be read with full_system/<record>/md.tpr.
• processed/<record>/traj.xtc must be read with processed/<record>/topology.pdb. Solvent, ions and hydrogens have been removed and the frames are fitted to the protein backbone.
• The two layers of one record share an MD run but not an atom set. Do not mix a topology from one layer with a trajectory from the other.

import mdtraj as md
t = md.load("processed/1ao7_run3/traj.xtc", top="processed/1ao7_run3/topology.pdb")

Simulation protocol

GROMACS 2023.2, CHARMM36m (charmm36-jul2021) with TIP3P water. Periodic cubic box, minimum solute-to-wall distance 1.0 nm, neutralized with Na+/Cl- to 0.15 M NaCl. Steepest-descent minimization, 100 ps restrained NVT at 310 K, 1 ns restrained NPT at 310 K and 1 bar, then 200 ns unrestrained NPT production with a 2 fs time step. PME electrostatics, 1.2 nm real-space cutoff, van der Waals force-switched from 1.0 to 1.2 nm, LINCS on hydrogen-containing bonds.

Replicas of a complex start from the same input structure. Most share one solvated system and differ only by independently randomized initial velocities, but a subset of complexes were solvated and ionized independently per replica, so their replicas differ in solvent atom count and box size as well. The solute is identical across the three replicas of every complex.

Intended use and limits

Each replica spans 200 ns and supports comparative characterization of docking geometry, peptide burial, contact occupancy and local flexibility within that sampling window. The trajectories do not resolve association or dissociation rates, rare conformational switching or force-dependent behaviour. The library is derived from the experimental structural record and inherits its composition biases, in particular the over-representation of HLA-A*02:01-restricted 9-mer complexes. A single force field was used throughout, which introduces a shared systematic bias.

Treat the three replicas of a complex as independent observations: compute metrics per replica before deriving complex-level summaries, and do not concatenate coordinate series without an analysis-specific rationale.

Source data

Structural inputs were obtained from the Protein Data Bank and annotated against the IMGT information system.

Licence

Released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0).