Propedia 26

Propedia 26 is a curated dataset of protein–peptide interactions derived from experimentally determined three-dimensional structures deposited in the Protein Data Bank (PDB). This record contains the complete set of structure files and the corresponding annotation table, allowing the dataset to be used offline and in large-scale computational workflows, including machine learning applications.

Each entry represents a protein–peptide pair: one chain of 2–50 amino acid residues (the peptide) in contact with one chain of more than 50 residues (the protein). Structures were retrieved from the PDB on 8 September 2025. A pair was retained when at least one atom of the peptide and one atom of the protein were separated by 6 Å or less and the complex showed a buried surface area greater than zero. This release comprises 73,392 protein–peptide complexes.

Files in this record

• propedia26_v8.zip (3.1 GB compressed; 15.4 GB uncompressed) — 73,392 structure files in PDB format, organised in directories named after the first character of the PDB identifier. Each file contains only the two chains that form the complex. File names follow the Propedia identifier convention <PDB ID>-<peptide chain>-<protein chain>, for example 1A1M-C-A.pdb.
• propedia26_v15.csv (61 MB) — annotation table with one row per complex (73,392 rows, 71 columns), semicolon-separated. The id column matches the structure file names.
• propedia26_removed_entries_bsa0.csv — 1,568 chain pairs from 126 PDB entries that satisfied the distance criterion but showed no buried surface area, and were therefore excluded from the released dataset. Provided for transparency and as a potential negative set for interface classification tasks.
• propedia26_v8.zip.sha256 — checksum for integrity verification.

Annotations

For every complex, the annotation table provides:

• Structure metadata retrieved from the PDB: title, classification, deposition date, resolution, experimental method, source organism, chain identifiers, descriptions and full sequences.
• Physicochemical properties computed separately for the peptide and the protein chain: molecular weight, isoelectric point, instability index, aliphatic index, GRAVY, hydrophobic percentage, numbers of positively and negatively charged residues, atomic formula, total atom count and molar extinction coefficients (with and without disulfide bonds). Calculated with ProtParam as implemented in Biopython, complemented by in-house scripts.
• Interface and surface descriptors: accessible surface area (ASA) of the complex, the peptide and the protein; buried surface area (BSA); buried peptide and protein areas; buried peptide percentage; and the list of protein residues within 6 Å of the peptide. Computed with NACCESS v2.1.1.
• Predicted interaction energy: binding affinity (kcal/mol) and dissociation constant (M) at 25 °C, together with interfacial contact counts by residue type and the percentage of apolar and charged non-interacting surface residues. Computed with PRODIGY v2.4.0.
• Predicted therapeutic profiles: probability of the peptide displaying anti-angiogenic (AAP), antibacterial (ABP), anticancer (ACP), anti-inflammatory (AIP), quorum-sensing (QSP) or surface-binding (SBP) activity.
• Clustering information: sequence, interface and binding-site cluster assignments, cluster leaders and 100% sequence identity groups.

Values in the therapeutic, energy and physicochemical fields are computationally predicted and should be interpreted as hypotheses for downstream screening, not as experimentally validated measurements. Binding affinities were obtained with a predictor originally trained on protein–protein complexes and are best used to rank complexes rather than as absolute values.

Important: extracting the archive

Chain identifiers in the PDB are case-sensitive, and large assemblies frequently use lowercase chain names. As a result, this dataset contains 4,174 files whose names differ from another file only by letter case, for example 1S5L-F-A.pdb and 1S5L-f-a.pdb, which are distinct complexes.

Extracting the archive on a case-insensitive file system — the default on macOS and Windows — will silently overwrite these files, leaving 4,174 structures missing and no warning message. Extract the archive on a case-sensitive file system (standard on Linux; on macOS, a case-sensitive disk image can be created with hdiutil). After extraction, confirm that the number of files is correct:

find pdb -name '*.pdb' | wc -l    # expected: 73392

Integrity verification

sha256sum -c propedia26_v8.zip.sha256

SHA-256 of propedia26_v8.zip:
68de0bcc0733902f4ad9c9f121af1519d498e1316497936d14638d3917e1bea0

Related resources

• Web interface, with interactive visualisation and search by sequence (BLAST) or by binding site (ProBiS): https://bioinfo.dcc.ufmg.br/propedia26
• Source code of the web tool: https://github.com/LBS-UFMG/propedia26
• Supplementary material and machine learning models: https://github.com/LBS-UFMG/propedia26-sm

License and attribution

The original structural data were obtained from the Protein Data Bank under the CC0 1.0 Universal Public Domain Dedication. The annotations and derived data distributed in this record are released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.