Pulling on a molecule: steered MD

The idea

Stop waiting,and pull

Ordinary MD watches a system do whatever it does. Steered MD attaches a spring to it and makes something happen on your schedule.

Rare events, on demand

Unbinding might take milliseconds unaided. A pull gets you there in nanoseconds of simulated time.

Force as an observable

The spring tension over the path is a direct, physical number you can plot and compare.

A route to free energy

Repeat the pull enough times and the spread of work values gives a free energy difference.

How it works

A spring,and a moving anchor

You pick a collective variable, attach a harmonic restraint to it, and move the restraint's centre at a constant rate.

The system resists. That resistance, recorded as the force in the spring over time, is the output you came for. Integrate it along the path and you have the work done on the system — and with enough repeats, Jarzynski's equality turns that distribution of work values into a free energy difference.

Pick the coordinateA distance, an angle, a centre-of-mass separation. Choosing it badly is choosing the wrong answer.

Move the anchorConstant velocity is the usual mode; constant force is the alternative when you want a fixed load.

Record the forceOne column of numbers per step, which is the whole result. The trajectory is supporting evidence.

What it answers

Four questionsworth pulling for

Steered MD is at its best where the interesting event is rare, directional, and something you can name a coordinate for.

Ligand unbinding

Pull a ligand out of a pocket and watch which contacts break, in what order, and against what force.

Protein unfolding

Stretch a chain and find the intermediates that a folding funnel drawing never shows you.

Membrane transport

Drag an ion or a small molecule through a channel and map the energetic cost along the pore.

Mechanical properties

Stiffness and elasticity of a molecule or a material, measured the same way you would measure a wire.

Honestly

Pull too fastand you measure the spring

The single most common mistake in steered MD is a pulling rate chosen for the compute budget rather than the physics.

SlowerPull ratethan the budget suggests

ManyRepeatsone pull is an anecdote

Halve itCheckif the answer moves, it is not converged

Pull quickly and the system has no time to relax, so the force you record is dominated by friction and the numbers come out far too large. Slow the pull down, run more repeats, and check that the work distribution is not still shifting. If halving the rate changes the answer, the answer is not converged yet — and that is a machine-hours problem, not a method problem.

GROMACS handles this with its pull code, AMBER through NMR-style restraints, and LAMMPS with fix smd. All three are here as workflows, so the choice is about which engine your system is already prepared for, not about what you can install.

On a machine

GROMACS isalready built

Nothing here needs compiling. Pick a machine, and these two commands put it on there.

hub://gromacs

$ dxflow workflow create --identity gromacs hub://gromacs

$ dxflow workflow start gromacs

Start a machine What GROMACS needs

Read it, then run itPrepaid and by the hour, on a machine that is yours about a minute after you ask.

Start a machine All notes