Skip to content
PaperFren

Research method

Molecular Dynamics Simulation

8 studies

Ask PaperFrenWatch this

Molecular dynamics integrates Newton's equations (classically, or with forces from DFT/QM/MM) to generate a trajectory of nuclear positions. Averages of distances, proton-transfer events, or metadynamics barriers are statistics of that trajectory at a chosen temperature and Hamiltonian — not a diffracted intensity or a measured rate. In this library MD ranges from AIMD of a MOF defect and of GFP's proton shuttle, through neural-network potentials trained on DFT water–gold data, to short classical MD of a calixarene–lysozyme crystal.

Chemists reach for MD when an experiment averages away the motion that might matter: a proton hopping on a missing linker, an excited-state shuttle in GFP, or an associative ORR path in water. It answers 'what does this model do over picoseconds to nanoseconds?' Its main limitation is that the dynamics are computed: they can be faster than femtosecond Raman, invisible to time-averaged XRD, and unable to rank binding sites even when a crystal structure exists.

Studies

8

Findings

5

6 supporting · 0 challenging · 0 qualifying citations

Open tensions

2

Currently

What we know

  1. AIMD of defective UiO-66 makes a proton shuttle that diffraction cannot see. Average Zr⋯O = 2.23 Å matches 300 K experiment, but a direct O0–O1 proton-transfer barrier is only 6.5 kJ mol⁻¹ versus 27.5 kJ mol⁻¹ for an O3-mediated path. NH₃ binds 110.1 versus 75.8 kJ mol⁻¹ at defect versus perfect regions; water can leave, exposing frustrated-Lewis-pair-like Zr. The dynamics are computed, not a measured catalytic turnover.
  2. A learned potential can push DFT-quality forces into nanosecond metadynamics. A PaiNN ensemble trained on PBE-D3 Au(100)–water data, then 2.5 ns path-CV metadynamics at 350 K, finds associative ORR: O₂ is hydrogenated toward *OOH, which is reduced to two *OH with neighbouring water rather than splitting into *OH + *O. The O₂-to-hydroxyl barrier is about 0.3 eV. The model is Au(100) in pure water without explicit cations or applied potential; the barrier is from the NNP landscape, not a measured Tafel slope.
  3. QM/MM AIMD of wtGFP in S₁ transfers a proton in four of five runs within 1 ps (720 fs in TRJI; 350, 380 and 980 fs in II–IV). O–O contacts shorten in 15–20 fs; twist modes drop from 146 to 101 cm⁻¹. Those computed times are faster than the experimental FSRS 3/10 ps phases (DFT barrier underestimation, no MM polarisation); nonadiabatic hops at longer times are omitted.
  4. MD plus docking can design peptides that then meet a wet IC₅₀, which is a stronger claim than MD that cannot rank sites. SARS-CoV-2 Mpro peptides p12–p16 have IC₅₀ 3.11–5.36 μM; at 8-fold excess they cut s01 depletion from >98% to 70–95%; the ion-pair dyad lies 28.5 kJ mol⁻¹ above the neutral state. On dimethylated lysozyme, short MD at each Lys-Me₂ and a crystal of sclx4 encapsulating Lys116-Me₂ still left MM-PBSA unable to rank sites; precipitation limited solution stoichiometry.
  5. TDDFT spectra computed on MD snapshots can assign a surprising UV-vis tail. Protein α3C (54% charged residues) has ε ≈ 7338 M⁻¹ cm⁻¹ at 250 nm and ε ≈ 964 and 501 M⁻¹ cm⁻¹ at 450 and 800 nm; 4–6 Å Lys–Glu contacts modulate ProCharTS charge-transfer bands. Computed CT energies inherit both the TDDFT functional and the MD ensemble.

Largest unresolved question

MD that is later tested in a cuvette is not the same deliverable as MD that stays inside the computer. Mpro peptides have micromolar IC₅₀ values; GFP AIMD proton-transfer times disagree with 3/10 ps FSRS phases; UiO-66 proton hops are invisible to time-averaged XRD. A student who says 'MD showed the mechanism' must say whether a wet observable confirmed it.

Procedure

How it works

The main stages of the method. Specific protocols and analysis choices vary between studies.

  1. Choose a Hamiltonian and a box

    Forces come from a classical field, DFT/QM/MM, or a neural-network potential trained on DFT. Temperature, solvent, and what is left out of the cell bound every later average; a missing polarisable term cannot appear later.

  2. Integrate Newton's equations

    Nuclear positions are stepped forward in time. The trajectory is the dynamics of this model over picoseconds to nanoseconds, not a diffracted intensity.

  3. Collect a trajectory

    Many frames are stored so that rare hops, distances, or a metadynamics path can be counted. A handful of AIMD runs is still a handful of runs.

  4. Average distances, hops, or barriers

    Papers report mean contacts, proton-transfer events, or a free-energy barrier. Matching a time-averaged XRD distance does not experimentally establish the hops the trajectory also contains.

  5. Treat the statistics as this model's motion, not a measured rate

    MD answers what this model does over the simulated window. Computed transfers can be faster than femtosecond Raman, invisible to diffraction, and unable to rank binding sites even when a crystal pose exists.

Common misconceptions

  • Molecular dynamics is a measurement of the real barrier or lifetime.

    GFP S₁ transfers in <1 ps in four trajectories, faster than 3/10 ps FSRS phases because DFT barriers are underestimated and MM is not polarisable. The Au ORR 0.3 eV barrier is an NNP/metadynamics landscape. UiO-66 6.5 kJ mol⁻¹ is a computed proton-transfer barrier, not a turnover frequency.

  • If the average MD distance matches XRD, the trajectory's proton hops are also experimentally established.

    UiO-66's 2.23 Å Zr⋯O matches 300 K diffraction; the 6.5 versus 27.5 kJ mol⁻¹ proton paths and NH₃ 110.1 versus 75.8 kJ mol⁻¹ binding are computed. Time-averaged XRD cannot see those hops.

  • Docking plus MD that produces an IC₅₀ means the computed protonation state is proven.

    Peptides p12–p16 inhibit at 3.11–5.36 μM; the 28.5 kJ mol⁻¹ ion-pair-over-neutral dyad preference remains a computed free energy. These are micromolar peptides, not oral antivirals.

Related