MD Trajectory Analysis

What is MD trajectory analysis?

MD trajectory analysis turns a simulation trajectory into measurements that can actually be interpreted: structural drift, residue flexibility, compactness, contact persistence, secondary-structure content, collective motions, and solvent-facing behavior. The raw frames coming out of GROMACS, AMBER, CHARMM, NAMD, or OpenMM do not answer those questions on their own.

This ProteinIQ tool runs a workflow pinned to MDAnalysis 2.11.0. It is not a single command-line program. It is a curated analysis surface built from MDAnalysis modules and a small amount of integration logic that standardizes inputs, runs multiple observables in one job, and returns structured outputs for charting and download.

The practical value is in combining observables that answer different failure modes:

• RMSD: Detects whether the selected atoms stay close to a chosen reference structure.
• RMSF: Shows which residues are mobile after aggregating atom-level fluctuations by residue.
• Radius of gyration: Tracks compaction or expansion over time.
• Distance tracking: Follows a specific geometric separation, often between domains, termini, ligands, or catalytic motifs.
• Hydrogen bonds and salt bridges: Measures persistence of stabilizing interactions.
• PCA and clustering: Describes dominant motions and recurring conformational states.
• DSSP, Ramachandran, chi angles, helix analysis: Connects motion to secondary structure and torsional states.
• SASA, RDF, density, water dynamics: Relates structural changes to solvent exposure and local environment.

Applications

• Fold stability checks: Compare early and late frames after simulations initialized from AlphaFold 2, Boltz-2, or ESMFold.
• Mutation analysis: Test whether a variant shifts flexibility, compaction, hydrogen-bond persistence, or helix geometry relative to the wild type.
• Binding-site monitoring: Track contact counts, hydrogen bonds, or a distance restraint around a ligand or interface during refinement.
• Ensemble characterization: Use PCA and clustering to summarize whether the trajectory samples one basin or several distinct states.
• Solvent-facing transitions: Use SASA, RDF, density, and water dynamics to examine collapse, exposure, or hydration changes.

How to use MD trajectory analysis online

Upload a topology file and a matching trajectory file, choose one or more analyses, and run the job to receive tabular and chart-ready outputs for each selected observable. The result page returns structural, interaction, conformational, environmental, and secondary-structure summaries with downloadable data for follow-up analysis.

Inputs

Input	Description	Accepted formats	Max size
Topology/Structure file	Defines atom identities, residue assignments, connectivity, and metadata needed to interpret trajectory coordinates.	.pdb, .ent, .cif, .gro, .crd, .psf, .prmtop, .top, .tpr	50 MB
Trajectory file	Stores coordinates for each sampled frame. The topology and trajectory must describe the same system and atom ordering.	.xtc, .trr, .dcd, .nc, .ncdf, .tng, .crd, .xyz, .mdcrd	50 MB

Common pairings include:

• .gro + .xtc for GROMACS
• .tpr + .xtc when topology metadata from GROMACS should be preserved
• .prmtop + .nc, .ncdf, or .mdcrd for AMBER
• .psf + .dcd for CHARMM or NAMD
• .pdb, .crd, or .gro with coordinate trajectories such as .xyz when atom ordering matches

How does MD trajectory analysis work?

The tool loads the uploaded files into an MDAnalysis Universe, applies the selected atom selection, resolves the reference mode, optionally aligns sampled frames to that reference, and then dispatches the requested observables.

Three details matter because they change interpretation:

• Reference mode: first, last, and average are not cosmetic options. With alignment on, average uses MDAnalysis AverageStructure, which fits the sampled frames to the first frame before averaging them.
• Frame sampling: step subsamples the trajectory globally for the audited analysis paths, which reduces runtime and directly changes temporal resolution.
• Alignment: When enabled, sampled frames are superimposed onto the chosen reference with the MDAnalysis QCP fit before RMSD, RMSF, PCA, clustering, and cross-correlation calculations. This removes rigid-body translation and rotation and makes those outputs more about internal motion than bulk drift.

Some analyses rely on additional assumptions from the underlying MDAnalysis modules or the tool:

• Hydrogen bonds: The tool runs MDAnalysis HydrogenBondAnalysis with its default donor-acceptor cutoff of 3.0 Å and angle cutoff of 150°, both adjustable. MDAnalysis infers hydrogens and acceptors from partial charges, which PDB and GRO topologies do not contain. For those files, set the H-bond hydrogen and acceptor selections, for example name H* and name O* N*. A topology without bonds, such as GRO, also needs the donor selection, for example name N* O*, so MDAnalysis can pair donors and hydrogens by distance. Without them the result shows the MDAnalysis error instead of a zero count. Explicit hydrogens are required either way.
• H-bond lifetime: Lifetime analysis uses the same MDAnalysis hydrogen-bond machinery and respects the selected atom group. A protein and name CA selection is useful for fold tracking but does not contain hydrogens, so lifetime analysis will usually return a warning unless a chemically complete selection is used.
• Contacts: Contact analysis uses MDAnalysis.analysis.contacts.Contacts and reports the source native-contact fraction Q. The displayed contact count is a compatibility summary derived from that fraction and the first-frame native contact set.
• MSD: Mean square displacement uses MDAnalysis.analysis.msd.EinsteinMSD with msd_type="xyz". Physical diffusion estimates require trajectories in an unwrapped coordinate convention.
• RMSF: MDAnalysis computes atom-level fluctuations. The tool then aggregates them by residue so the output is residue-oriented even when the selection contains more than one atom per residue.
• DSSP: Secondary-structure assignment uses MDAnalysis.analysis.dssp on every atom of the residues in the atom selection, so a name CA selection still provides complete backbones. It reports helix (H), strand (E), and loop (-) codes. If DSSP cannot run, for example with fewer than six residues, the result shows the MDAnalysis error.
• Salt bridges: Within the residues touched by the atom selection, a charged residue pair (Arg, Lys, or His against Asp or Glu) counts once per frame when any of its charged side-chain atoms are within 4 Å, using the periodic box when the trajectory has one. Occupancy is the share of sampled frames in which the pair is bridged.
• SASA: Surface area comes from FreeSASA with its default atom radii. Hydrogens are left out, as in FreeSASA's own structure input.
• Chi angles and helix geometry: Side-chain χ1/χ2 summaries use MDAnalysis Janin analysis with circular angle statistics. Helix geometry uses MDAnalysis HELANAL on alpha-carbon selections and requires consecutive helix segments of at least 9 residues.
• Water dynamics: This mode is available in the current tool, but it depends on water naming conventions and native MDAnalysis functionality that is being deprecated on the path to MDAnalysis 3.0.

Understanding the settings

Core settings

Setting	Description	Default
Analyses to run	Selects which observables are calculated. Multiple analyses can run in one job.	RMSD, RMSF, Radius of gyration
Atom selection	MDAnalysis selection string applied to the primary analysis group. Examples include protein, protein and name CA, backbone, and resid 1-100.	protein and name CA
Reference frame	Reference for RMSD-aligned analyses. Average structure computes the mean coordinate set over the sampled frames.	first

Advanced settings

Setting	Description	Default
Frame step	Samples every nth frame. A value of 10 processes frames 0, 10, 20, ... instead of every saved frame.	1
Align trajectory	Aligns sampled frames to the chosen reference before RMSD, RMSF, PCA, clustering, and cross-correlation analysis.	true
Second atom selection	Secondary MDAnalysis selection used for distance tracking and RDF.	resname SOL and name OW
H-bond donor selection	MDAnalysis donors_sel. Blank infers donors from the hydrogens.	blank
H-bond hydrogen selection	MDAnalysis hydrogens_sel. Blank infers hydrogens from masses and partial charges.	blank
H-bond acceptor selection	MDAnalysis acceptors_sel. Blank infers acceptors from partial charges.	blank
H-bond donor-acceptor cutoff	MDAnalysis d_a_cutoff, the maximum donor-acceptor distance.	3.0 Å
H-bond angle cutoff	MDAnalysis d_h_a_angle_cutoff, the minimum donor-hydrogen-acceptor angle.	150°
PCA components	Number of principal components returned in PCA projections.	3
Clusters	Number of k-means clusters used for conformational clustering.	5
Density axis	Axis used for one-dimensional density profiles.	z

Selection examples

Selection string	Meaning
protein	All atoms in standard amino-acid residues
protein and name CA	One alpha carbon per residue, often used for coarse-grained structural tracking
backbone	Backbone atoms N, CA, C, O
resid 45-70	Residues 45 through 70
protein and not name H*	Heavy atoms only
resname ATP	All atoms in ATP residues, useful for ligand-focused measurements

Understanding the results

The result page groups outputs into tabs. Not every selected analysis appears in every tab, so the most useful way to read the output is by question rather than by file format.

Structural outputs

Analysis	Output fields	What it answers
RMSD	frame, time_ps, rmsd_angstrom	Is the selected structure drifting away from the reference?
RMSF	residue_id, residue_name, chain, rmsf_angstrom	Which residues fluctuate the most over the sampled frames?
Radius of gyration	frame, time_ps, rg_angstrom	Is the selected structure compacting or expanding?
End-to-end distance	frame, time_ps, distance_angstrom	Are termini or endpoints moving apart or together?
Asphericity	frame, time_ps, asphericity	Is the shape becoming more elongated or more spherical?
Moment of inertia	frame, time_ps, I1, I2, I3	How is mass distribution changing along principal axes?

Interaction outputs

Analysis	Output fields	What it answers
Hydrogen bonds	frame, time_ps, hbond_count; pair table with donor, acceptor, occupancy	Are stabilizing hydrogen-bond networks persistent?
Salt bridges	frame, time_ps, salt_bridge_count; pair table with residue ids and occupancy	Are charged interactions forming or breaking over time?
Contacts	frame, time_ps, native_contacts_q, n_contacts	Is an interface retaining its first-frame native contacts?
Distance tracking	frame, time_ps, distance_angstrom	Is a specific geometric feature moving toward or away from another one?
H-bond lifetime	lifetime distribution and summary statistics	Are hydrogen bonds brief, recurrent events or long-lived contacts?

Conformational outputs

Analysis	Output fields	What it answers
PCA	frame, time_ps, pc1, pc2, pc3 plus explained variance arrays	What are the dominant collective motions?
Clustering	per-frame cluster assignments and cluster statistics	How many recurring conformational states are populated?
MSD	MDAnalysis EinsteinMSD time series and diffusion estimate	Is there translational diffusion or large-scale wandering?
Dynamic cross-correlation	significant residue pairs and correlation coefficients	Which residue motions are coupled or anticorrelated?
Persistence length	estimated persistence length and correlation decay	How stiff is a chain-like segment across the sampled trajectory?

Secondary structure and environment outputs

Analysis	Output fields	What it answers
DSSP	per-frame helix, sheet, and coil percentages	Is secondary structure stable, gained, or lost?
Ramachandran	residue-level mean and standard deviation for phi and psi	Which residues move into strained or alternative backbone states?
Chi angles	Janin χ1/χ2 circular means and standard deviations	Are rotamers stable or switching between states?
Helix analysis	HELANAL rise, twist, bend, tilt, and variability summaries	Are helices remaining canonical or distorting?
SASA	per-frame total SASA and per-residue SASA summary	Is the structure exposing or burying surface area?
RDF	radial bins and g(r) values	How is local solvent or particle organization distributed around a selection?
Density	position bins and density values along one axis	Is mass or solvent redistributed along a membrane or box axis?
Water dynamics	residence-time summary statistics	How long do hydration waters remain near the selected region?

Interpreting results

RMSD and RMSF

RMSD is easiest to misread when alignment and atom selection are not considered together.

• Low, stable RMSD after alignment usually indicates internal structural stability of the selected atoms.
• Low RMSD without alignment disabled can hide large rigid-body motion if the entire selection rotates or translates together.
• High RMSF in loops or termini is often expected.
• High RMSF inside a buried core helix or beta strand usually deserves inspection, especially when it appears alongside SASA growth or DSSP loss.

Distance, contacts, and hydrogen bonds

Distance tracking is strongest when paired with contacts or hydrogen bonds.

• Distance increases while contact count drops often indicates real separation rather than internal breathing.
• Stable distance with fluctuating contacts can mean the interface remains nearby but repacks locally.
• Hydrogen bonds near zero with an explicit warning usually means the uploaded structure lacks hydrogens, not necessarily that the system has no hydrogen bonding in reality.

PCA and clustering

PCA answers whether motion is dominated by a few directions. Clustering answers whether those motions produce recurring states.

• High PC1 variance with one dominant cluster suggests one major breathing or hinge motion inside one broad state.
• Several populated clusters with separated PC projections suggests discrete conformations rather than one continuous fluctuation cloud.
• Poorly separated clusters often mean the trajectory is either short, noisy, or not naturally state-separated under the chosen atom selection.

DSSP, SASA, and compactness

These metrics become more informative when read together.

• Rg decreases while SASA decreases usually indicates compaction.
• DSSP helix or sheet content falls while RMSF rises suggests local unfolding or disordering.
• SASA rises without major Rg change often reflects side-chain or surface rearrangement rather than global unfolding.

When to use MD trajectory analysis vs alternatives

MD trajectory analysis on ProteinIQ is a post-simulation interpretation tool. It is most useful after a trajectory already exists.

Tool	Best use case	Not the right choice when
MD Trajectory Analysis	Summarizing structure, interaction, and solvent observables from an existing MD run	No trajectory exists yet
OpenMM	Running or setting up molecular dynamics simulations	The simulation is already complete and only analysis is needed
MDGen	Generating conformational samples when no trajectory exists yet	An experimental or simulated trajectory already exists
gmx_MMPBSA	Estimating binding free energy from MD snapshots	The goal is structural interpretation rather than free-energy estimation
Ramachandran Plot	Inspecting backbone geometry from a single structure	Time-resolved trajectory behavior is needed
RMSD Calculator	Comparing one structure against another	A full time series across many frames is needed

The main decision point is whether the question is dynamic or static:

• For one structure or one pairwise comparison, a simpler structure-analysis tool is usually enough.
• For questions about persistence, transitions, convergence, flexibility, or hydration over time, trajectory analysis is the correct layer.

Caveats that matter in practice

• Sampling frequency still limits interpretation: A frame saved every 10 ns cannot recover fast side-chain switching, no matter how many downstream analyses are run.
• Selection choice changes the science: protein and name CA is useful for global fold tracking, but it can miss side-chain rearrangements and interaction chemistry.
• Hydrogen-dependent analyses depend on protonation and topology quality: Missing hydrogens or inconsistent naming can suppress hydrogen-bond signals.
• MSD assumes unwrapped coordinates: Wrapped trajectories can make diffusion estimates physically misleading even when the calculation succeeds.
• HELANAL requires alpha-carbon helix segments: Short selections or non-consecutive residue groups may return a warning instead of helix geometry.
• Subsampling trades speed for detail: Larger step values reduce runtime and noise, but short-lived events may disappear.
• Water dynamics is not future-proof: The current tool still supports it, but the underlying MDAnalysis path is being deprecated on the route to version 3.0.