hazemessam/fireprot_db

FireProtDB 1.x Export

Protein mutations with measured effects on stability (ΔΔG and/or ΔTm), as exported from FireProtDB 1.x (Stourac et al., Nucleic Acids Research, 2021) in the file fireprotdb_results.csv.

FireProtDB has since been replaced by FireProtDB 2.0, and the 1.x export can no longer be downloaded from https://loschmidt.chemi.muni.cz/fireprotdb/. This file is byte-identical to the copy of the 1.x export published in skalyaanamoorthy/PSLMs (same SHA-256 as that repository's Git LFS object).

Quick start

from datasets import load_dataset

ds = load_dataset("hazemessam/fireprot_db", split="data")
df = ds.to_pandas().drop_duplicates()  # see "Duplicate rows" below

Contents

53,445 rows and 35 columns, as exported by FireProtDB.

Duplicate rows. Only 15,987 rows are distinct: the other 37,458 repeat an earlier row exactly, in all 35 columns. Each experiment_id corresponds to exactly one distinct row. The duplicates are part of the export itself; drop them before training or evaluating.

Among the distinct rows, 12,131 have ddG and 3,879 have dTm.

Sign of ddG. Positive ddG means the mutation destabilizes the protein: 73% of the distinct ddG values are positive, and ddG is strongly anti-correlated with dTm (Pearson −0.85, where negative dTm is a lower melting temperature). This is the opposite of PROSTATA's ddg.

Columns

FireProtDB's column names, unchanged:

column	meaning
experiment_id	identifier of the measurement
protein_name, uniprot_id, pdb_id, chain	the protein and structure
position, wild_type, mutation	the substitution
ddG, dTm	labels: change in folding free energy and in melting temperature; either can be empty
sequence	protein sequence; empty in some rows
method, method_details, technique, technique_details, pH, tm, notes	experimental conditions
publication_doi, publication_pubmed	source publication
is_curated, datasets	FireProtDB curation flag and the datasets the record belongs to
other columns	structural and evolutionary annotations computed by FireProtDB: type, derived_type, interpro_families, conservation, is_essential, correlated_positions, is_back_to_consensus, secondary_structure, asa, is_in_catalytic_pocket, is_in_tunnel_bottleneck, b_factor, hsw_job_id

Verification

Checked on 2026-11-13 against the original source. SHA-256 of each file in this repo:

file	SHA-256	check against source
fireprotdb_results.csv	f24c84e02bd39f6941d6cf0a78a174f2b85a96cc0f7441ca85d2cc23994e36be	byte-identical to the copy in skalyaanamoorthy/PSLMs (official 1.x download no longer available)

Source, license and citation

FireProtDB, https://loschmidt.chemi.muni.cz/fireprotdb/. FireProtDB's paper calls the database freely available but states no license for the data, so the license here is unknown. Please cite FireProtDB:

@article{stourac2021fireprotdb,
  title   = {FireProtDB: database of manually curated protein stability data},
  author  = {Stourac, Jan and Dubrava, Juraj and Musil, Milos and Horackova, Jana and Damborsky, Jiri and Mazurenko, Stanislav and Bednar, David},
  journal = {Nucleic Acids Research},
  volume  = {49},
  number  = {D1},
  pages   = {D319--D324},
  year    = {2021},
  doi     = {10.1093/nar/gkaa981}
}

Changes

README written and configs: added on 2026-11-13 (previous revision d4f09a5). The license was changed from MIT to unknown, since FireProtDB states no license. Data file unchanged.