Tutorial 5 — Slab simulation of phase separation (LLPS)

Goal: measure whether — and at what temperature — an IDP undergoes liquid–liquid phase separation (LLPS) using the standard slab method: pack many copies of a chain into an elongated box so the system separates into a dense (condensate) and a dilute (dispersed) phase coexisting across a flat interface. This is the workhorse setup of the IDP condensate field.

Time: the demo md_npt.ini leg is short, but slab systems are large (thousands of beads, many chains) — a GPU is strongly recommended for real runs. On CPU the demo is slow; treat it as a mechanism walk-through.

Files in this folder

File

Role

NON.pdb

A multi-chain starting system (many copies of the IDP in one box).

md_npt.ini

Stage 1: NPT compression to a sensible density.

run_simulation.py

Thin runner wrapper.

The slab method, in three stages

You do not know the right box size a priori, so you build it in stages (staged build protocol):

  1. Start big, then compress (NPT). Put all chains in a generously large cubic box and run NPT (pcoupl = yes) so the Monte Carlo barostat shrinks the box to a reasonable, condensed density. This is md_npt.ini.

  2. Elongate one axis. Take the equilibrated x = y box edge from stage 1, keep x and y, and extend z several-fold (e.g. [L, L, 3L]). The slab of protein now occupies the middle of a long box with empty space on either side along z.

  3. Hold volume, let it separate (NVT). Turn the barostat off (pcoupl = no, fixed box) and run a long NVT simulation. The chains redistribute into a dense slab coexisting with a dilute vapor; the interfaces are perpendicular to z.

Step-by-step

1. Stage 1 — NPT compression

python run_simulation.py -f md_npt.ini

Watch the box dimensions in the build/run log shrink as the barostat drives the system toward ref_p = 1 bar at ref_t = 310 K. Outputs go to traj/NON.*.

2. Stage 2 — build the elongated box (manual)

Read the final box edge from stage 1 and make a stage-2 config that keeps x,y and stretches z, e.g.:

pbc = yes
box_dimension = [L, L, 3L]   ; substitute the equilibrated L from stage 1
pcoupl = no                  ; NVT from here on
restart = yes                ; continue from traj/NON.chk

(You start stage 2 from traj/NON_final.pdb / traj/NON.chk.)

3. Stage 3 — production NVT & read off the phase diagram

Run a long NVT simulation, then compute the density profile along z, ρ(z). Phase separation shows up as a flat-topped high-density plateau (the condensate) bracketed by a near-zero baseline (the dilute phase). The two plateau values are the coexisting densities at that temperature.

Repeat the whole procedure across a ladder of ref_t values: as you raise the temperature the dense and dilute densities approach each other and finally merge at the critical temperature — that curve is the model’s phase diagram.

Background: reading phase behavior off a slab

  • Why a slab (elongated box)? The flat interface minimizes curvature/finite- size artifacts, so the two coexisting densities are well defined and converge quickly compared with a droplet geometry.

  • Concentration & temperature are the knobs. A single slab at one temperature gives one tie-line (two densities). Scanning temperature traces the binodal; the system phase-separates below the critical point and stays mixed above it.

  • Model choice matters. The driving force for separation comes straight from the hydropathy parameters (Tutorial 2). mpipi is tuned for near-quantitative phase behavior; the hps family is the classic choice. Compare against the literature for whichever you pick.

Try next

  • Run stage 1 on a GPU (device = GPU) for a realistic system size.

  • Build the stage-2/stage-3 configs and compute ρ(z) with MDAnalysis (histogram bead positions along z, average over the trajectory).

  • Move to Tutorial 6 for a multi-component condensate: protein + RNA.