CABS-flex 3.0 Tutorial: Protein RMSF Without GROMACS
Skip to content

CABS-flex 3.0 Tutorial: Protein RMSF Without GROMACS

CABS-flex 3.0 Tutorial: Protein RMSF Without GROMACS

CABS-flex 3.0 is a free web server (lcbio.pl/cabsflex3) that simulates protein flexibility with a coarse-grained model instead of all-atom molecular dynamics. You give it a PDB ID or upload a PDB/mmCIF file of up to 2,000 residues and 10 chains, choose a flexibility mode, and it returns a per-residue RMSF plot, 10 all-atom models, contact maps and the trajectory, with no login and no GPU.

This tutorial walks through one complete job on a real structure, explains which of the four flexibility modes to pick, shows how to read every results tab, and is honest about where a CABS-flex RMSF stops being a substitute for a GROMACS run. It is written by the StemSkills Lab team, who have spent more than ten years in structural bioinformatics, drug design and multiscale molecular modelling.

What is CABS-flex and how is it different from molecular dynamics?

CABS-flex is a protein flexibility simulator built on the CABS coarse-grained model, developed by Sebastian Kmiecik’s group at the University of Warsaw. Instead of every atom, each residue is represented by up to four pseudo-atoms (the C-alpha, the C-beta, the side-chain centre and the peptide-bond centre), and the simulation uses Monte Carlo sampling rather than integrating Newton’s equations step by step. At the end, the coarse-grained models are rebuilt to all-atom structures.

The current version is described in Wróblewski et al., “CABS-flex 3.0: an online tool for simulating protein structural flexibility and peptide modeling”, Nucleic Acids Research 53(W1):W95-W101 (2025). The authors summarise it in one line: “CABS-flex 3.0 enables efficient simulations of protein and peptide dynamics.” The paper states the method is three to four orders of magnitude faster than all-atom MD, which is why a job that would need days of GPU time in GROMACS finishes on a shared web server.

The original method was published as CABS-flex in Nucleic Acids Research 41(W1):W427 (2013), so if your reviewer asks “is this validated?”, you can cite a method with more than a decade of published use.

When should a student use CABS-flex instead of GROMACS?

Use CABS-flex when the question is “which regions of this protein are flexible?” and you do not need solvent, ions, a ligand, or a time axis in nanoseconds. Typical good uses:

  • A quick flexibility profile of an AlphaFold or homology model before you commit compute to a full MD run.
  • Comparing the flexibility of a wild-type protein with a point mutant in a short project.
  • Deciding which loops to treat carefully in homology model refinement or in flexible docking.
  • Students with no GPU, no HPC account and a thesis deadline.

Use GROMACS when you need protein-ligand stability, hydrogen bonds over time, binding free energy, a membrane, or any claim about real time scales. CABS-flex has no explicit water and no ligand handling, so it cannot answer those questions. Our GROMACS learning path covers that workflow end to end.

MethodWhat it samplesLigand and solventTypical costBest for
CABS-flex 3.0Coarse-grained Monte Carlo, rebuilt to all-atomNo ligand, implicit environmentWeb server job, no hardwarePer-residue flexibility profile, quick screening
iMODS (normal mode analysis)Collective motions around one structureNoWeb server job, no hardwareLarge-scale hinge and domain motions
GROMACS all-atom MDTime-resolved trajectory of every atomExplicit water, ions, ligands, membranesHours to days on a GPUComplex stability, interactions, energetics

If you want to compare CABS-flex with an elastic network approach, see our iMODS normal mode analysis tutorial.

How do you run a CABS-flex 3.0 job step by step?

We will use ubiquitin, PDB 1UBQ (X-ray, 1.8 Å, 76 residues, one chain), because it is small, well studied and finishes fast.

  1. Open the server. Go to lcbio.pl/cabsflex3. There is no registration. The site states it is free and open to all users with no login requirement.
  2. Give the structure. Type the four-character ID 1UBQ in the PDB code box, or click Browse and upload your own PDB or mmCIF file. The structure is the only required input.
  3. Name the project. Optional. If you leave it blank, the server assigns a random hash code, which is hard to find later, so type something like ubq_flexible_T1.4.
  4. Click “Show options”. Here you choose the flexibility mode (Rigid-pLDDT, Rigid, Flexible, Unleashed), the chains to simulate, the pLDDT source, an email for notification, and a checkbox to keep your job off the public results page. Tick that box for unpublished thesis work.
  5. Pick the mode. For a well-folded crystal structure like 1UBQ, start with Flexible and run a second job in Rigid (see the next section).
  6. Leave advanced settings alone on your first run. The documented defaults include a simulation temperature of 1.4 (in CABS units, not kelvin), a restraint gap of 3, and restraint lengths from 3.8 to 11.5 Å.
  7. Click Run and keep the results URL. That link is your job record.

Want the guided, hands-on version?

Our live Molecular Modeling & MD Simulations cohort bootcamp takes you from zero to running real docking and MD workflows, with a portfolio project for your grad-school applications.

Join the waitlist (free) →

Which flexibility mode should you choose: Flexible, Rigid, Rigid-pLDDT or Unleashed?

The mode controls the distance restraints that hold the starting structure together during the simulation. This is the single most important choice in a CABS-flex job, and it is the one students most often leave unreported in their methods section.

ModeWhat the restraints doUse it when
FlexibleRestraints only on residues in secondary structure elements, so loops and disordered regions move freelyYou want to see loop and terminus mobility in a folded protein
RigidUniform restraints on all residues, minimising fluctuationsStable, well-folded proteins; the paper reports this mode correlated highly with ATLAS MD fluctuation profiles for globular proteins
Rigid-pLDDTRestraint strength scaled by per-residue pLDDT and secondary structureAlphaFold models; pLDDT is read from the B-factor column or a .json/.tsv file
UnleashedNo restraints, only the CABS force fieldExploratory work only; the authors note it agrees less with MD and may exaggerate motions

For an AlphaFold model, Rigid-pLDDT is the natural choice because it lets low-confidence regions move and keeps confident ones in place. If you are unsure what those scores mean, read our guide to AlphaFold pLDDT and PAE first. The paper notes Rigid-pLDDT usually gives flexibility closer to Rigid than to Flexible.

A reasonable protocol for a thesis chapter is to run the same structure in two modes and report both. If a loop is flexible in Rigid as well as Flexible, that finding is much harder for a reviewer to dismiss.

How do you read the CABS-flex results?

When the job finishes, the results page has four tabs.

Project information

This tab is available immediately and lists the rigidity, restraint and simulation parameters actually used. Screenshot or copy it. Your methods section needs the mode, the temperature and the number of cycles, and this is where they are recorded.

Models

The Models tab shows the starting structure with flexibility colouring (RMSF, pLDDT or B-factor), each of the 10 final models one at a time, and all 10 superposed. The “All models” view is the most useful figure for a slide: rigid regions overlap tightly and flexible regions fan out. A tar.gz archive of the models is available for download, and you can open them in PyMOL with the steps in our superposition guide.

Contact maps

An interactive map of intra- and inter-chain residue contacts. Hovering shows the residue pair and contact frequency across the simulation, and clicking a contact highlights those residues in the 3D view. For a multi-chain input this is where you check whether an interface stays formed.

Fluctuation plot

This is the figure most students want: RMSF on the y-axis against residue number on the x-axis, with secondary structure annotated and an optional pLDDT trace. You can download the values as CSV and replot them in your own style. For 1UBQ, check whether the highest peak sits at the C-terminal tail (the sequence ends in LRLRGG at residues 71 to 76), the region a reader of the ubiquitin literature would expect to move most.

Read the plot relatively, not absolutely. Peaks tell you which residues are more mobile than others in this model. The absolute height depends on the temperature and the restraint mode, so a CABS-flex RMSF of 3 Å is not directly comparable with a 3 Å RMSF from a 100 ns GROMACS run.

How does a CABS-flex RMSF compare with a GROMACS RMSF?

If you have both, compare the shape of the profiles, not the numbers. In GROMACS, a per-residue RMSF comes from:

gmx rmsf -s md.tpr -f md.xtc -o rmsf.xvg -res

Two practical points before you overlay them:

  • Units differ. GROMACS writes RMSF in nanometres; the CABS-flex plot is in ångströms. Multiply the GROMACS values by 10.
  • Residue numbering must match. If your GROMACS topology renumbered residues from 1, map them back to the PDB numbering before plotting.

Then compute a simple Pearson correlation between the two columns, and state the result in words: “CABS-flex reproduced the flexible regions seen in MD (loops X and Y) but under-estimated the motion of the N-terminus.” That sentence is more defensible than a figure with two curves and no comment. Our RMSD and RMSF tutorial for GROMACS covers the MD side.

Be careful with one claim in particular. The CABS-flex 3.0 validation used the ATLAS MD database (nearly 1,400 proteins), which the authors note favours stable, well-folded monomers. For intrinsically disordered proteins or large multi-domain systems, do not assume the same agreement holds.

Why does my CABS-flex job fail, and how do I fix it?

The server pages do not list error messages, so these are the input problems that follow from its documented limits, with the fix for each.

  • Structure too large. The limit is 2,000 residues and 10 chains. Use the chain selection option to simulate only the chains you care about, or trim distant domains before upload.
  • Ligands and waters in the file. CABS-flex simulates proteins and peptides. Remove HETATM records for ligands, ions and water before uploading, for example in PyMOL with remove hetatm followed by save clean.pdb. The ligand would not be part of the simulation anyway.
  • Missing residues or chain breaks. Gaps in the backbone give misleading flexibility at the break. Model the missing loop first; our MODELLER tutorial shows one route.
  • Rigid-pLDDT chosen with no pLDDT. An experimental PDB file carries crystallographic B-factors, not pLDDT. Use Rigid-pLDDT only for AlphaFold models, or upload the scores as a .json or .tsv file.
  • Unrealistic motions in Unleashed mode. This is expected behaviour, not a bug. Rerun in Flexible or Rigid for any figure you plan to publish.

For anything else, the developers point users to the CABS-flex GitHub issue tracker.

What should you write in your methods section?

A reviewer needs enough to repeat the job. Report the server and version, the input structure (PDB ID or model source), the chains simulated, the flexibility mode, the temperature and number of cycles from the Project information tab, and the citation. For example: “Protein flexibility was simulated with the CABS-flex 3.0 web server (Wróblewski et al., 2025) using PDB 1UBQ chain A in Flexible mode at the default temperature of 1.4; per-residue RMSF was taken from the 10 final models.”

If CABS-flex is one step in a larger modelling project, our computational biology skills roadmap shows where flexibility analysis sits between structure prediction, docking and MD.

Frequently asked questions

Is CABS-flex 3.0 free?

Yes. The web server at lcbio.pl/cabsflex3 is free, open to all users and has no login requirement. An email address is optional and only needed if you want a notification when the job ends.

Can CABS-flex simulate a protein-ligand complex?

No. CABS-flex models proteins and peptides. Remove the ligand before uploading. For protein-ligand stability you need all-atom MD, for example in GROMACS.

How many models does CABS-flex produce?

Each job returns 10 final all-atom models, plus the trajectory, contact map data and the RMSF values as a downloadable CSV.

Can I use CABS-flex on an AlphaFold model?

Yes. Choose Rigid-pLDDT mode, which reads pLDDT from the B-factor column of an AlphaFold PDB file and lets low-confidence regions move more.

Is a CABS-flex RMSF enough for a thesis, or do I still need MD?

It is enough for a flexibility profile and for choosing regions to study further. It is not enough for claims about ligand binding, interaction lifetimes or time scales. Present it as a fast complementary method and say so in your text.

What temperature does CABS-flex use?

The documented default is 1.4 in CABS reduced units. The help page suggests 1.6 or 2.0 when you want to see larger motions. Report whichever value you used.

Want the guided, hands-on version?

Our live Molecular Modeling & MD Simulations cohort bootcamp takes you from zero to running real docking and MD workflows, with a portfolio project for your grad-school applications.

Join the waitlist (free) →

Think you know Molecular Dynamics (GROMACS)?
Take the free StemSkills assessment and earn a verifiable certificate you can download and add to your LinkedIn profile.
Start the free assessment

Keep going

TM-align Tutorial: Compare Protein Structures by TM-score Install TM-align, compare a predicted model with an experimental structure, read both TM-scores correctly and see why RMSD… Periodic Image Check in GROMACS: gmx mindist -pi Prove your protein never touched its own periodic image: run gmx mindist -pi, read the xvg against your… mdpocket Tutorial: Pocket Volume Over a GROMACS Trajectory Check if your binding pocket stayed open during MD: prepare a fitted GROMACS trajectory, run mdpocket twice and…
See live workshops