Independent exploratory project, built with AI-assisted learning and troubleshooting (see Acknowledgments).
An AutoDock Vina pipeline that docks a polyamide dimer into NylC, a nylon-oligomer hydrolase, comparing the isolated monomer with the native tetrameric assembly. It is a first, hypothesis-generating step before molecular dynamics (MD) simulations of the catalytic mechanism. It is not a validated binding study.
Status: exploratory phase completed in September 2026.
Polyamides (nylons) are an abundant class of plastics whose recycling remains difficult (Bourgery et al., 2026). Depolymerizing them back into reusable monomers with enzymes is therefore of interest for sustainable plastics recycling. NylC is a nylon-oligomer hydrolase of the nylonase family, recognized for promising enzymatic properties in polymer degradation (Sun et al., 2025). The structure used here is PDB 3AXG, NylC from Agromyces sp. strain KY5R.
NylC is produced as a precursor that cleaves itself. The new N-terminal residue created by this autoproteolytic cleavage, Thr-267, is the catalytic residue, and its side-chain oxygen (OG1) acts as the nucleophile that attacks the amide bond (Negoro et al., 2012). In 3AXG, the disordered remnant of the cleavage site (residues 261 to 266) is unresolved in most chains, while Thr-267 is resolved in all chains used here. The asymmetric unit contains 15 identical chains (A to O); chains A, B, C and D form the biological tetramer.
The project was carried out in 2026 in the context of a collaboration with Dr. Dirk Wacker (Polysecure GmbH), outside any formal internship. It is a preliminary study for a larger project that aims to stabilize NylC together with its polyamide substrate in a non-natural medium. The docking study had two goals:
- Identify candidate binding poses near Thr-267 that could serve as starting points for MD simulations.
- Check whether docking into the native tetramer changes the outcome compared with the isolated monomer, which tests whether the oligomeric context needs to be modelled in later steps.
Because rigid-receptor docking has well-known limits, the results below are meant to generate hypotheses and to flag methodological pitfalls, not to establish binding preferences.
nylc-docking/
├── data/
│ ├── raw/ # Unmodified inputs: 3AXG.pdb, FASTA, PA2_opt.pdb
│ └── processed/ # Derived files: chain A extract, cleaned tetramer, PDBQT conversions
├── scripts/ # Pipeline: structure prep, docking, log parsing, replicates
├── config/ # Vina parameters per run (monomer and tetramer)
├── results/
│ ├── tables/ # Parsed scores and replicate summaries (CSV)
│ ├── figures/ # Pose visualizations, in mono/ and tetra/ subfolders
│ └── docked_poses/ # Docked structures (PDBQT)
└── docs/ # Extended methods notes
- Receptor preparation. Chain A (monomer) or chains A to D (tetramer) are extracted with PyMOL, with solvent and sodium ions removed, then converted to PDBQT with MGLTools.
- Ligand preparation. The polyamide dimer
PA2_opt(about 40 atoms, 13 rotatable bonds, one internal hydrolyzable amide bond) is converted to PDBQT with MGLTools. - Docking. AutoDock Vina 1.2.7 with a 25 Å cubic search box centered on Thr-267, exhaustiveness 24, 9 modes, identical for both receptors.
- Replicates. Each condition is run with 5 explicit seeds to measure run-to-run variation of the mode 1 score.
- Pose analysis. Poses are inspected in PyMOL, including the distance between Thr-267 OG1 and the ligand's hydrolyzable carbonyl carbon.
Open Babel was first tried for the PDBQT conversion, but its PDBQT output plugin did not work on this installation, so all PDBQT files were generated with MGLTools. Full details (grid box rationale, inter-chain distance checks, tool quirks) are in docs/methods.md.
- AutoDock Vina 1.2.7
- MGLTools 1.5.7
- PyMOL
- Python 3.10+ (standard library only, see
requirements.txt)
The Python scripts are thin wrappers: they call Vina, MGLTools and PyMOL, run the replicates and parse the Vina logs into CSV tables. The docking itself is done by Vina.
Edit scripts/paths_config.py to match your local install locations for MGLTools and Vina. This is the only file that needs machine-specific changes.
Note: Windows install paths were tested directly. The macOS and Linux paths in
paths_config.pyare typical defaults and have not been verified.
python scripts/prepare_receptor.py monomer
python scripts/prepare_receptor.py tetramer
python scripts/prepare_ligand.py
python scripts/run_docking.py monomer
python scripts/run_docking.py tetramerEach run_docking.py call runs Vina and parses the log into results/tables/{label}_scores.csv.
To reproduce the replicate analysis:
python scripts/run_replicates.py monomer 1 2 3 4 5
python scripts/run_replicates.py tetramer 1 2 3 4 5Monomer. In the default run, the top-scoring pose (mode 1) scores −5.496 kcal/mol, with 9 modes within a 4 kcal/mol window. Modes 2 and 3 lie close to mode 1 (RMSD lower bound of about 1.7 to 2.0 Å), while modes 4 to 8 diverge more (RMSD upper bounds above 6 Å for some), consistent with the ligand's flexibility (13 rotatable bonds). Over 5 seeds, the mode 1 score is −5.338 ± 0.183 kcal/mol (mean ± SD).
Tetramer. In the default run, the top-scoring pose scores −5.735 kcal/mol. Over 5 seeds, the mode 1 score is −5.907 ± 0.177 kcal/mol.
Monomer versus tetramer. The tetramer scores about 0.57 kcal/mol better on average, and this difference is reproducible across seeds (the seed-to-seed ranges do not overlap: −5.02 to −5.47 for the monomer, −5.70 to −6.12 for the tetramer). However, it is smaller than the typical uncertainty of Vina scoring (of the order of 2 to 3 kcal/mol), and the protocol has not been validated by re-docking a known ligand. The difference is therefore not interpreted as evidence that the tetrameric context improves binding.
Possible contribution of the search box. Widening the residue shell around chain A's Thr-267 to 6 to 8 Å found no residue from chains B, C or D, so the immediate pocket does not appear to be altered by the other chains. However, 34 atoms from chains B, C and D lie within 13 Å of Thr-267 OG1, which is inside the 25 Å search box. These atoms may contribute to the small score difference, but this was not tested (for example by repeating the monomer run with the same atoms present, or the tetramer run with a trimmed box), so it remains a hypothesis.
Catalytic geometry (OG1 to carbonyl carbon distance). These distances were measured once, on the poses of the default run, and not on every replicate.
| Mode 1 | Mode 2 | Mode 3 | |
|---|---|---|---|
| Monomer | 3.1 Å | 4.3 Å | 5.2 Å |
| Tetramer | 4.8 Å | 5.8 Å | 3.8 Å |
Score and geometry rank poses differently. The monomer's best-scoring pose (mode 1) also has the shortest distance, but the tetramer's best-scoring pose (mode 1) does not: tetramer mode 3, third by score, has the most reaction-plausible geometry. This illustrates that Vina's score alone does not identify a reaction-competent pose, so geometric inspection relative to Thr-267 is needed in both conditions.
Full tables are in results/tables/, and pose visualizations are in results/figures/mono/ and results/figures/tetra/.
- No validation of the protocol. No re-docking of a known ligand was performed, so the accuracy of Vina's poses and scores for this system is unknown. This is the first control to add.
- Small score difference. The monomer versus tetramer gap is reproducible but below the typical scoring uncertainty, so it should not be read as a biological effect.
- Search box composition. Atoms from other chains inside the box may influence the scores. Grid box design should be scrutinized whenever receptors with different overall geometry are compared.
- Rigid receptor. NylC is treated as rigid. Side-chain and backbone flexibility around the active site are not captured, and protonation states use standard residue defaults rather than a pKa prediction.
- Score versus reactive geometry. Vina's scoring function has no term for catalytic geometry, which is why poses were inspected relative to Thr-267 OG1.
- No experimental data. Nothing here has been compared with activity or binding measurements.
- Next. MD simulations of the most reaction-plausible candidates (monomer mode 1 and tetramer mode 3) to assess pose stability, preceded by the re-docking control.
This project was completed independently by the author, with AI assistance (Claude by Anthropic) used as a learning and troubleshooting aid: explaining docking and force-field concepts as they arose, debugging pipeline scripts, and structuring this repository. All docking runs, parameter choices, PyMOL geometric analyses and scientific interpretation are the author's own work.
- Bourgery et al., 2026
- Negoro et al., 2012
- Sun et al., 2025
This project is licensed under the MIT License. If referencing this work, please cite: https://github.com/darni3l/nylc-docking