Protein Stability · LiteFold
FireProtDB
FireProtDB — curated thermostability mutation data for proteins, supporting protein-engineering and stability-prediction modelling.
What FireProtDB is
FireProtDB is an extensive repository of experimentally determined protein thermostability and fitness data. Thermostable proteins are vital across biomedical and industrial biotechnology, yet laboratory-based mutagenesis campaigns to improve stability are resource-intensive. Computational stability prediction addresses this challenge, but models require vast, systematic experimental datasets for robust training and validation.
The database encompasses wild-type sequences, mutants, isolated protein domains, and de novo designed proteins, capturing both single-point mutations and complex alterations such as insertions, deletions, and multi-point substitutions.
The dataset provides standardised thermodynamic parameters, assay conditions, sequence annotations, and structural descriptors. By capturing experimental metrics alongside AlphaFold and Protein Data Bank structural context, it serves as an empirical foundation for training machine learning models in protein engineering and computational biophysics.
How it was made
The dataset aggregates stability measurements from primary experimental initiatives, notably Domainome, MegaScale, ProTherm, and COZYME. It establishes an abstract schema that harmonises diverse measurement records into a structured 99-column flat table. Data types cover both relative and absolute stability indicators, including melting temperature changes (dTm), unfolding free energy changes (ddG), enthalpy (dH), and heat capacity (dCp), with secondary measurements accessible in raw JSON formats.
Data records are assigned fixed identifiers and partitioned into deterministic training and test splits via a cryptographic hash of source row indices (SHA-256 modulo 10). Split 0 is reserved for test sets (546,499 rows), while buckets 1 through 9 comprise the training split (4,919,161 rows). Tabular data are distributed in Apache Parquet format alongside structural, sequence, and experimental condition metadata.
Key facts
- Total rows
- 5,465,660
- Mutant rows
- 5,453,252
- Sequence rows
- 12,408
- Total columns
- 99
- Data format
- Parquet
- Primary source datasets
- Domainome DDG, MegaScale, Domainome FITNESS, ProTherm, COZYME
- Mutation event counts
- 5,543,778 substitutions, 54,296 deletions, 50,613 insertions
- Train/test split size
- 4,919,161 train / 546,499 test
What you can use it for
- Machine learning for stability prediction
- Train and benchmark predictive deep learning models that infer changes in protein melting temperature (dTm) and free energy of unfolding (ddG) directly from sequence or structural data.
- High-throughput mutagenesis analysis
- Evaluate the biophysical effects of specific insertions, deletions, and multi-point substitutions on target protein scaffolds during protein engineering campaigns.
- Contextual stability investigation
- Assess how environmental variables such as pH, ionic concentration, and temperature modulate experimental protein stability and unfolding pathways.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from LiteFold/FireProtDB at the moment you asked. Nothing is cached or stored — every row above came from that request.
Strengths and limits
Good at
- Standardises diverse biophysical readouts including dTm, ddG, dH, dCp, and protease resistance alongside primary assay conditions.
- Includes deterministic train and test splits generated by row-level cryptographic hashing for reproducible machine learning benchmarks.
- Links sequence variants directly to three-dimensional structural descriptors from the Protein Data Bank (PDB) and AlphaFold DB.
Watch out for
- Scalar columns store only the first measurement when multiple experimental replicates exist, requiring users to parse raw JSON fields for complete histories.
Questions
Can I look inside FireProtDB without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is FireProtDB under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
What types of protein mutations are included in the dataset?
The database includes single-point amino acid substitutions, multi-point substitutions, insertions, and deletions across natural proteins, protein domains, and de novo designed constructs.
Which biophysical stability measurements are reported?
Primary scalar metrics include melting temperature (tm, dtm), free energy changes (dg, ddg), transition enthalpy (dh, dhvh), heat capacity (dcp), denaturant midpoints (cm, m_value), and protease stability (trypsin_ml, chymotrypsin_ml).
How are the training and test splits constructed?
Splits are assigned deterministically using the formula sha256('fireprotdb:{row_index}') modulo 10. Row bucket 0 is assigned to the test set, while buckets 1 to 9 are placed in the training set.
Does the dataset include structural structural information?
Yes, records are mapped to structure coordinates from the Protein Data Bank (PDB) and AlphaFold Database (AFDB), alongside structural annotations such as secondary structure, accessible surface area (ASA), B-factors, and pocket or tunnel presence.
How to cite it
Musil, M., Borko, S., Planas-Iglesias, J., Lacko, D., Rosinska, M., Kabourek, P., Martins, L. O., Tataruch, M., Damborsky, J., Mazurenko, S., & Bednar, D. (2026). FireProtDB 2.0: large-scale manually curated database of the protein stability data. Nucleic Acids Research, 54(D1), D409–D418. https://doi.org/10.1093/nar/gkaf1211
Written from FireProtDB's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.