Tutorial A.7 — A protein with intrinsically disordered regions (IDRs)

Goal: simulate a protein that is part folded, part disordered, and see the disordered regions behave as flexible, non-folding chains while the folded core keeps its structure. This is the only new ingredient beyond Tutorial A.2: a disordered: section in domain.yaml.

Time: ~2 minutes on a CPU for the 100 000-step demo (365 residues).

Prerequisite: do Tutorial A.1 and Tutorial A.2 first — this tutorial reuses the same runner and the same domain.yaml file, adding one section.

The system

P28566 (5-HT1E receptor, 365 residues) modelled from its AlphaFold structure. We treat it as one folded domain plus two intrinsically disordered regions (IDRs):

  • folded core — most of the chain (domain A), held together by native contacts;

  • N-terminal tail (residues 1–18) — disordered;

  • long intracellular loop (residues 218–281) — disordered.

An IDR is not a fold, so its equilibrium shape is set by the potential, not by the AlphaFold coordinates: TOPO removes the native contacts of the disordered residues and replaces them with a weak, non-specific attraction, so those regions relax into a flexible ensemble while the core stays folded.

Files in this folder

File

Role

P28566.pdb

Input structure (AlphaFold model; TOPO keeps only the CA atoms).

domain.yaml

Folded domain A plus a disordered: section marking the IDRs.

md.ini

Simulation configuration (steps, temperature, I/O, hardware).

P28566_stride.dat

Precomputed STRIDE H-bonds (so the run does not call STRIDE).

The one new thing — the disordered: section

Open domain.yaml. Compared with a normal folded domain.yaml, the only addition is the disordered: block:

n_residues: 365
intra_domains:
  A:
    residues: [18-164, 177-219, 279-363]
    nscale: 2.5044
disordered:
  residues: [1-18, 218-281]     # N-terminal tail + long intracellular loop
  idr_scale: 0.10               # IDR-IDR well depth, x eps_BT (default)
  eps_ev_kj: 0.8368             # IDR repulsive core, kJ/mol (default)

  • residues — the residues to treat as disordered. Their native contacts (H-bond, backbone–sidechain, and sidechain–sidechain) are removed.

  • idr_scale — the scale on the sequence-dependent BT energy, which varies by residue-pair type (optional, default 0.10, calibrated against SAXS radii of gyration for 18 disordered proteins). This is the compaction knob: raising it lowers the scaling exponent ν, as attraction physically should. Set it to 0 for a pure self-avoiding chain. The θ point is at ≈ 0.32, so treat anything much above ~0.3 as suspect.

  • eps_ev_kj — the repulsive-core strength in kJ/mol of the Ashbaugh–Hatch potential used for IDR-involving pairs (optional, default 0.8368, the HPS value). It sets bead size, independently of the well depth — which is the point of that functional form, and why idr_scale: 0 now gives a self-avoiding chain of physical thickness rather than a near-phantom one.

    Disordered↔folded pairs take the same depth rule as disordered↔disordered pairs — the interaction depends on the two residue types, not on which region each one is in — so an IDR can transiently associate with the domain surface. What it never regains is a native (fold-encoding) contact. Note idr_scale was calibrated on fully-disordered chains, so the cross-pair depth is an extrapolation.

You choose the disordered residues. TOPO applies exactly the set you list — it does not detect disorder. Sources to inform the choice include MobiDB and the AlphaFold pLDDT < 70 proxy (read from the B-factor column). See the full write-up: Disordered / IDR regions.

Overlap is allowed — disorder wins. Here domain A and the IDR ranges touch at a few residues (18, 218–219, 279–281). The run prints an info line (IDR overlap: residues [...] are in both a domain and 'disordered:'; treating as disordered.) and those residues are treated as disordered. This is a feature: you can define a domain broadly and carve a disordered loop out of it.

Run it

topo-mdrun -f md.ini            # == python -m topo.mdrun -f md.ini

Early in the build you will see the overlap info line and the contact count. With the IDRs masked, P28566 has 823 native contacts (the disordered residues contribute none). Outputs land in traj/ exactly as in Tutorial A.1.

The disordered regions start off the model’s energy minimum (their native contacts are gone), so they carry some force at step 0 and relax over the first steps. This is expected — discard the initial relaxation in analysis, as for any MD run. (minimize = no; see the note in md.ini.)

What to look for

The folded core keeps its shape while the tail and loop swing freely. Aligning each frame on the folded core and measuring the per-residue fluctuation (RMSF) over the 100-frame demo makes this quantitative:

Region

mean RMSF

Folded core (19–164)

~0.7 Å

N-terminal IDR (1–18)

~17 Å

Intracellular-loop IDR (218–281)

~26 Å

i.e. the disordered regions are ~20–35× more mobile than the core — flexible, but still attached and (at the default attraction) compact rather than fully extended. Watch it in VMD:

vmd traj/traj.psf traj/traj.dcd

Try next

  • Self-avoiding vs collapsing. Set idr_scale: 0 in domain.yaml and rerun; the IDRs expand further. Compare the radius of gyration of a disordered region against the default run.

  • A fully disordered protein. Drop intra_domains and list every residue under disordered: — a whole-IDP run (see the fully-IDP note in the Disordered / IDR regions page).

  • Calibrate the core. The nscale = 2.5044 here is a placeholder; use the optimizer (Tutorial A.5) to calibrate it with the disordered: section active, so the folded core is tuned under the real (IDR-reduced) energy.

  • Longer sampling. The demo is short; a converged IDR ensemble needs far more steps (the validation run used md_steps = 1_000_000) and is best on a GPU (device = GPU).

Full theory and the exact energy/well-position rules for each pair class (folded–folded, IDR–IDR, IDR–folded) are in Disordered / IDR regions.