Package: pdbmdauto 1.2.0 · Template id: pdbmdauto-pipeline

Install the package from the Marketplace, then open this template from Start from Template. Its inputs ship with the package or are fetched for you — the walkthrough says which, and whether it needs the network. The bundled template takes one PDB identifier and returns a complete, protonated, solvated and energy-relaxed system with a short test trajectory — every intermediate on disk.

Before you run

• Network. Step 1 fetches 4Z8J from the RCSB PDB. Offline, set Input mode to a local file.
• A working directory. Everything lands under <working directory>/pdbmdauto-e2e-full/e2e_4z8j/ (the case name comes from the template). Choose an empty folder.
• The first install builds a 3.1 GB environment once (GROMACS, ProMod3, OpenStructure, PDB2PQR, PROPKA, Biopython, OpenMM). On Windows Salpa sets that environment up inside WSL2 for you.

The structure

The model after step 5. The five-residue tag on the domain’s N-terminus and the peptide’s first residue were not located in the experiment; ProMod3 built them. They hang off the ends, where a terminal extension has nothing to pack against — a reminder that a rebuilt residue is a prediction.

The pipeline

Eleven nodes in a line. Each reads what the previous one wrote and adds its own files to the case directory. Paths below are relative to pdbmdauto-e2e-full/e2e_4z8j/.

1 · PDB FASTA Parser — Biopython, RCSB PDB

Downloads the entry, splits it by chain, and reads the header’s REMARK 465 records into one CSV per chain: which residues the sequence has and the coordinates do not. For 4Z8J that is five rows for chain A (33–37) and one for chain B (586).

2 · Generate Alignment

Aligns each chain’s resolved residues to its full sequence — the gaps in that alignment are what step 5 will build.

3 · Multi-Chain Alignment

Consolidates the per-chain alignments into one, so the complex is modelled as a unit.

4 · Merge PDB Chains — Biopython

Writes the selected chains as one model, dropping water and hetero residues (ligands, ions, cofactors). If your system needs a bound ligand, this is the step to notice.

5 · Fix Missing Residues — ProMod3 / OpenStructure

Builds the six missing residues, reconstructs sidechains and briefly minimises. In the reference run chain A went from 96 to 101 residues and chain B from 7 to 8. Two things to know: ProMod3 renumbers from 1 (merge.pdb carried 38–133 and 587–593; fixed.pdb reads 1–101 and 1–8), and the Model Terminal Extensions option has no effect in this version — termini are always built.

6 · pKa + GROMACS EM — PROPKA, PDB2PQR, GROMACS

pdb2pqr --ff AMBER --titration-state-method=propka --with-ph=7.00 decides which His, Asp, Glu, Lys and Arg are charged at pH 7 and adds hydrogens; pdb2gmx -ff amber99sb -water tip3p -ignh writes the topology; editconf -d 2 puts a box around it; two steepest-descent minimisations follow, first without constraints, then with H-bond constraints. Reference run: the constrained one converged to Fmax < 1000 in 15 steps at −1.26 × 10⁴ kJ/mol.

7 · Original Atom Groups — GROMACS

Marks which atoms came from the experiment, so the next step can hold them still while the rebuilt ones move. Reference run: OriHeavy 776 atoms, OriBackBone 414 — the 103 resolved residues × 4 backbone atoms + 2 terminal oxygens, exactly the six rebuilt residues excluded.

8 · GMX MD Relaxation (restrained) — GROMACS, protocol full_4step

Four stages in vacuo: 5 ps of NVT at 300 K with every experimental heavy atom frozen, the same with only the experimental backbone frozen, then two conjugate-gradient minimisations. The rebuilt residues settle; the crystal structure does not drift. Reference run: the last minimisation converged to Fmax < 100 in 7 steps at −1.55 × 10⁴ kJ/mol. This is the slowest step — 84 s on a cold laptop, 222 s on the same laptop once hot.

9 · GMX Solvate & Ionize — GROMACS

editconf -box 5 5 5, solvate with SPC216, then genion -neutral -conc 0.15 -pname NA -nname CL. Reference run: 3,509 waters; the complex carried a charge of −1, so 12 Na⁺ and 11 Cl⁻ bring it to zero at 0.15 M; 12,193 atoms in total.

10 · GMX MD Relaxation (solvated) — GROMACS, protocol em_only

One steepest-descent minimisation of the whole box, 500 steps, tolerance 500. Reference run: it stops at the step limit with Fmax ≈ 2.9 × 10³ kJ mol⁻¹ nm⁻¹ and a potential energy of −1.84 × 10⁵ kJ/mol — the water has relaxed onto the protein (compare the curve below) but the tolerance is not met. For the demonstration that is enough; for a production study raise nsteps in em.mdp until it converges.

11 · GROMACS MD Run (Local) — GROMACS

1000 steps of 2 fs at 300 K (velocity rescaling, no pressure coupling), coordinates every 100 steps: 11 frames in md.trr (uncompressed; no .xtc is written). gen_seed = -1 in the shipped md.mdp, so velocities — and therefore trajectories — differ from run to run. Reference run: 31.6 ns/day on the cold laptop, 9.1 ns/day once hot.

Looking at the result

vmd Merge/fixed.pdb                      # the completed model — colour by chain, find the termini
vmd gmx/md.gro gmx/md.trr                # the trajectory, 11 frames
printf 'Potential\n0\n' | gmx energy -f gmx/em.edr -o em.xvg   # the curve above
grep -E 'converged|Potential Energy|Maximum force' gmx/em_hbonds.log gmx/em.log

Expected result

From the reference run (Intel MacBook Pro, Salpa 0.3.1, pdbmdauto 1.2.0, 2026-09-03):

rebuilt residues     A 33-37 (GSHGG) + B 586 (Gln)  →  fixed.pdb chain A 1-101, chain B 1-8
index groups         OriHeavy 776   OriBackBone 414   Protein 1643 (with H)
vacuum EM (step 6)   converged Fmax < 1000 in 15 steps   E = -1.26e4 kJ/mol
solvated system      12193 atoms   3509 SOL   12 NA   11 CL   charge -1 → 0   box 5.0 nm
solvated EM (step 10) 500 steps, not converged   E = -1.84e5 kJ/mol   Fmax = 2.9e3
production           1000 × 2 fs = 2 ps   11 frames in md.trr   9-32 ns/day

What varies between runs: the rebuilt residues’ exact coordinates (ProMod3 samples), the MD velocities (random seed), and the last digits of the energies. What must not vary: six residues rebuilt, 12,193 atoms, a neutral system, eleven nodes green.

Runtime

The Intel spread is thermal: a back-to-back run on a hot laptop took 2.5× longer in every GROMACS stage. The template’s “5 minutes” is the honest round figure for an Intel laptop.

Regenerating the figures

vmd/README.md beside this file — two headless VMD renders and one matplotlib plot, from the files a run leaves behind.

Watch