MacroCycle Dataset: From PDB Structures to a Computation-Ready Benchmark

MacroCycle Dataset is a curated collection of experimentally determined protein–macrocycle binding modes for evaluating macrocycle conformational sampling, structure preparation, docking, and related molecular-modelling methods.

Release summary

Item	Value
Protein–macrocycle binding-mode cases	1,861
Unique unprotonated macrocycle structures	1,570
Unique PDB entries	1,686
Primary macrocycle size	8–40 atoms

This is the first frozen public release. Its acceptance criterion is a genuine, single-component macrocycle with no known hard error in the unprotonated ligand structure. The pH 7.4 protonated derivative is supplied for computational convenience and is not used as an acceptance criterion.

Repository layout

MacroCycleDataset/
├── README.md
├── LICENSE
├── CITATION.cff
├── benchmark_index.csv
├── assets/
│   └── dataset_overview.jpg
└── cases/
    └── {case_id}/
        ├── protein.pdb
        ├── ligand_unprotonated.sdf
        └── ligand_protonated.sdf

Files in each case

File	Description
protein.pdb	Protein structure supplied as the experimental binding context.
ligand_unprotonated.sdf	Curated experimental ligand coordinates with audited atom connectivity and bond orders, before pH-dependent protonation. This is the reference ligand structure for benchmark evaluation.
ligand_protonated.sdf	Computationally prepared pH 7.4 form with explicit hydrogens, generated from the curated unprotonated structure while retaining its heavy-atom coordinates; high-confidence acid/base errors were subsequently audited and corrected.

Core case metadata and the unprotonated/protonated isomeric SMILES are consolidated in benchmark_index.csv.

Dataset index

benchmark_index.csv is the canonical one-row-per-case index. It contains exactly 1,861 records and uses relative case identifiers rather than machine-specific paths.

Column	Meaning
case_id	Stable case directory name.
pdb_id	RCSB PDB accession.
resolution_A	Experimental structure resolution in ångströms, when reported.
structure_title	Deposited PDB structure title.
publication_doi	DOI of the associated primary publication, when available.
unprotonated_isomeric_smiles	Isomeric SMILES derived from the curated reference ligand.
protonated_isomeric_smiles	Isomeric SMILES for the computationally prepared protonated form.
primary_ring_size	Number of atoms in the primary macrocycle. Ordinary small rings are not counted as macrocycles.

Missing source metadata are represented by empty CSV fields rather than invented values.

Dataset construction

Macrocycle candidates were collected by screening experimentally determined structures in the RCSB PDB rather than relying only on keywords or ligand annotations. LigandExplorer (Li et al., 2026) was used to separate candidate ligands from protein complexes, after which residue- and atom-level connectivity was analyzed to confirm genuine macrocycles, remove non-target or structurally unusable entries, and retain the protein chains representing the observed binding context.

For each accepted case, experimental heavy-atom coordinates were preserved while chemical connectivity and bond orders were reconstructed with Guess Bond Order and checked, where available, against the original PDB/mmCIF record, CCD definitions, deposited instance-coordinate files, and primary literature. The curated unprotonated SDF serves as the structural reference; a pH 7.4 protonated SDF was generated as a computation-ready derivative and audited for high-confidence protonation errors. Automated validation and manual PyMOL review were used to ensure consistency among the ligand structures, SMILES, metadata, and protein–ligand pairing before freezing the release.

Quick start

Inspect the index:

import pandas as pd

index = pd.read_csv("benchmark_index.csv")
print(index.shape)                 # (1861, 8)
print(index.primary_ring_size.describe())

Load a reference ligand with RDKit:

from pathlib import Path
from rdkit import Chem

case_id = "10hy__Ligand_organic_A1C5F_A_301"
path = Path("cases") / case_id / "ligand_unprotonated.sdf"
mol = Chem.SDMolSupplier(str(path), removeHs=False)[0]
assert mol is not None

Citation

Please cite this release using the metadata in CITATION.cff and cite the original experimental structure publications through the DOI information in benchmark_index.csv whenever individual cases are discussed.

License and source attribution

The benchmark compilation, curation, index, documentation, and original figure are distributed under CC BY 4.0. Structural data derived from the PDB archive remain available under the wwPDB/RCSB PDB CC0 policy. See LICENSE for details.

• RCSB PDB usage policy: https://www.rcsb.org/pages/usage-policy
• CC BY 4.0: https://creativecommons.org/licenses/by/4.0/

The data are provided without warranty. Users remain responsible for checking fitness for their intended scientific application.