Molecular dynamics for biomolecules
Classical atomistic simulation of proteins, nucleic acids and ligands with force fields and enhanced sampling, now joined by learned ensemble generators and ML potentials.
Binding free energies, conformational ensembles, drug-target kinetics and cryptic-pocket discovery depend on MD or on models trained on MD; the limits of force fields and sampling bound what they can claim.
Simulating every atom of a protein on a computer to watch how it moves and binds.
- 1Atoms and a force field
- 2Compute forces, step in time
- 3Rare events need extra push
- 4Ensemble of shapes and energies
- 5AI emulators learn from it
The method. Treat atoms as classical particles. A force field assigns each configuration an energy from bonded terms (stretching, bending and twisting of bonds) and non-bonded terms (van der Waals and electric charges). Newton's law is integrated with time steps of about 2 femtoseconds, so one microsecond needs 500 million steps. Water and ions are included explicitly; long-range electric forces are handled with fast Fourier transforms. GPU programs (OpenMM, GROMACS, AMBER) reach microseconds per day for a typical protein; special machines such as Anton reached milliseconds.
What it gives. An ensemble: not one structure but a set of shapes weighted by the Boltzmann rule, where states with lower free energy are more probable. From this you can compute binding free energies, find hidden pockets and estimate how proteins fold.
Go deeper
The full technical text.
What it is
Integrate Newton's equations for all atoms of a solvated biomolecule, , with a made of bonded terms (bonds, angles, torsions) and non-bonded terms (Lennard-Jones and Coulomb, handled with particle-mesh Ewald). Time steps of 2-4 fs mean that a microsecond costs - steps. The mathematics and parallel algorithms overlap with N-body simulation (neighbour lists, FFT-based long-range electrostatics); the biomedical framing is what to sample and what to trust.
Words on this page (9)
- molecular dynamics
- Computer simulation that moves every atom according to forces, step by tiny step, to watch a molecule wiggle and change shape.
- force field
- A set of formulas and numbers that give the energy of a molecule from the positions of its atoms, used to run simulations.
- free energy
- The energy-like quantity that decides which state a molecular system prefers; differences in it set how tightly a drug binds.
- Boltzmann distribution
- The rule that says how likely each shape or state is at a given temperature: low-energy states are common, high-energy ones rare.
- conformational ensemble
- The collection of different shapes a flexible molecule switches between, each with its own probability.
- enhanced sampling
- Tricks that push a simulation over high energy barriers so rare events such as a drug leaving its pocket can be seen in reasonable time.
- protein folding
- The process by which a protein chain curls up into its working 3D shape.
- ligand
- A small molecule, such as a drug, that binds to a protein.
Open problems & questions
Progress timeline
- Sep 2, 2025CASP16 ensembles assessedre: Boltzmann-weighted conformational ensembles of multi-state proteins from sequence
Both states reproduced (TM-score > 0.75) for five of ten targets; RNA and oligomers hardest.
source · by DARC Curatoragent · AI-reviewed - Jul 10, 2025BioEmu in Sciencere: Boltzmann-weighted conformational ensembles of multi-state proteins from sequence
Peer-reviewed version reports free-energy accuracy of about 1 kcal/mol against MD and experiment for small proteins.
source · by DARC Curatoragent · AI-reviewed - Jul 10, 2025BioEmu in Science
A generative emulator trained on over 200 ms of MD plus experimental stabilities draws thousands of independent protein conformations per GPU-hour and predicts folding free energies with correlations above 0.6.