TrajMine

Point it at a molecular-dynamics trajectory and get back a mechanism + kinetics report — tICA, a validated Markov State Model, macrostates, MFPTs and contact-map diffs — without hand-building the MSM.

checking backend…

For MD / biophysics labs GROMACS · OpenMM · AMBER · NAMD Force-field developers Turnkey · opinionated · one report

Run the MSM demo

Alanine dipeptide, 250 ns — the canonical MSM benchmark. Runs the full pipeline server-side on CPU and returns a live mechanism report.

⚠ The live server isn't active right now. The hosted demo is on-demand and may be offline.

See the static example report · source on GitHub · or run it locally on your own trajectory (below).

Mechanism report — alanine dipeptide, 250 ns open in new tab ↗

Run it on your own trajectoryrecommended

Real trajectories are usually multi-GB, so they're run locally via the CLI — chunked I/O means they never load fully into RAM, and nothing leaves your machine. Handles .xtc / .trr / .dcd / .nc / .h5 / .pdb / .gro.

python trajmine.py analyze traj.xtc --top top.pdb

Kinetics

tICA slow coordinates → reversible MSM → implied-timescales convergence → PCCA+ macrostates → stationary populations → mean first-passage times.

Mechanism

Per-macrostate contact maps and the contact-map difference between the two dominant states — the residue contacts that form or break across the transition.

Geometry & structure

Backbone RMSD(t), per-residue RMSF, PCA landscape, Ramachandran, native-contact fraction Q(t), DSSP composition + per-frame timeline.

TrajMine starts where the simulation ends and stops at the mechanism report. Built on deeptime (kinetics) and mdtraj (structure); methods after Pérez-Hernández 2013 (tICA), Prinz 2011 (MSM), Röblitz–Weber 2013 (PCCA+), Best–Hummer–Eaton 2013 (native contacts). CPU-only, no GPU. · github.com/gianyrox/trajmine · product brief · example report