Protein Microenvironment Data for Structure Learning with Convolutional Neural Networks

17263 mmCIF files representing 3D protein crystal structures and 1,455,978 JSON objects identifying unique protein microenvironments. The dataset includes four trained TensorFlow models and was authored by James Loy, last updated on March 18, 2024. Each JSON object identifies a microenvironment using a PDB code, chain ID, residue sequence number, and the amino acid present.

Use Cases

• Training convolutional neural networks for protein structure prediction based on the provided 3D mmCIF files.
• Analyzing local protein microenvironments based on the 1.4 million JSON objects with PDB codes and residue identifiers.
• Benchmarking machine learning models for structural biology tasks using the four included trained TensorFlow models.
• Studying the relationship between amino acid sequence and local 3D structure based on the microenvironment annotations.

Strengths

• Contains 1,455,978 annotated protein microenvironments, providing a substantial basis for analysis.
• Includes 17,263 individual 3D protein structure files in mmCIF format.
• Provides four pre-trained TensorFlow models, offering a potential starting point for related tasks.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the JSON objects is known, but the structure and scale of the mmCIF file archives are not detailed.
• Description metadata is limited; actual data quality and model performance require manual inspection after download.

Provenance

Source

Texas Data Repository Harvested Dataverse

Collection Method

Likely compiled from the Protein Data Bank (PDB) based on the use of PDB codes and mmCIF formats.

Time Range

null

Freshness

Last updated 2024-03-18 05:06:18; freshness should be verified.

Geography

null