FireProtDB-PDB: Protein Stability Changes from 3,438 Mutations

3,438 point mutations across 100 proteins are linked to their experimentally measured thermodynamic stability changes (ddG) and corresponding wild-type protein structures from the RCSB Protein Data Bank. This dataset was compiled for training models like ThermoMPNN to predict the effect of mutations on protein stability. It includes curated PDB files and CSV data with splits for training, validation, and homologue-free testing.

Use Cases

• Training machine learning models to predict ddG values based on protein structure and mutation data.
• Benchmarking protein stability prediction algorithms using predefined train/val/test splits.
• Studying the relationship between single amino acid substitutions and thermodynamic stability across diverse proteins.

Strengths

• Contains 3,438 experimentally measured stability changes (ddG) for point mutations, a specific quantitative resource.
• Integrates sequence-based measurements from FireProtDB with curated 3D structural data from the RCSB PDB.
• Includes predefined dataset splits (train/val/test/homologue-free) used in the original ThermoMPNN study for reproducibility.

Limitations

• The dataset size (3,438 mutations) is relatively small for modern deep learning applications.
• Metadata such as exact row count, file size, and last update date are not provided across the sources.
• The 0-based pdb_position index may differ from the original sequence position, requiring careful mapping.

Provenance

Source

FireProtDB (sequence-based ddG measurements) and RCSB Protein Data Bank (structures).

Collection Method

Data was assembled by matching sequence-based ddG measurements in FireProtDB to structures from the PDB.