Bitbol-Lab/ProteomeLM-ess-data

ProteomeLM gene-essentiality data

The data used to train and evaluate the gene-essentiality classifier of ProteomeLM (PNAS 2026, Fig. 5; trained head: Bitbol-Lab/ProteomeLM-ess). The essentiality calls come from OGEE v3, the Online GEne Essentiality database (Gurumayum et al., NAR 2021). The OGEE website (v3.ogee.info) is currently unavailable, so its tables are mirrored here unmodified under its CC BY 3.0 license. Please cite OGEE, and the primary screens it compiles, when you use these labels.

Files

file	content
raw/ogee_v3/gene_essentiality.txt	OGEE essentiality calls (dataset, taxaID, locus, gene, score, essentiality, pmid, Ref_db)
raw/ogee_v3/genes.txt	OGEE gene annotations
raw/ogee_v3/datasets.txt	OGEE datasets (source, technique, condition, definition of essential)
data/proteins.parquet	486,681 proteins of 89 genomes, with labels and folds (config proteins)
data/genomes.tsv	one row per genome: organism, protein and label counts, role
data/minimal_cells.parquet	JCVI-Syn1.0 and JCVI-Syn3A proteomes and labels (config minimal_cells)
data/info.json	split parameters

proteins

Each OGEE genome's proteome (UniProt reference proteome, else UniProtKB, else NCBI or the source named in OGEE) with exact duplicates (on the first 4,096 residues) removed. OGEE calls were matched to proteins by gene name, locus tag and synonyms.

column
taxid	NCBI taxonomy id of the genome
protein_id, header	FASTA id and full header line
sequence	amino-acid sequence
gene, synonyms	gene name and synonyms used for matching
ogee_calls	all OGEE calls for the gene (E essential, NE non-essential, C conditional, ...)
label	training label: the last E/NE call, null if none
fold	cross-validation fold: 0 = test, 1 = validation, 2-4 = train
fasta_file	proteome file name in the original pipeline

153,918 proteins have a label (42,077 E, 111,841 NE). 82 genomes form the cross-validation set; the others were not used for training:

taxid	organism	role	proteins	E	NE
83333	Escherichia coli K-12	held out (Fig. 5B)	4,361	290	3,969
199310	Escherichia coli CFT073	excluded	5,336	232	0
580240	Saccharomyces cerevisiae S288C	held out (Fig. 5B)	5,975	939	4,353
679895	Escherichia coli BW25113	excluded	4,192	313	0
941322	Escherichia coli O25b:H4-ST131	excluded	4,821	294	4,023
1124478	Vibrio cholerae O1 str. C6706	excluded	3,532	0	0
1380365	Dyella japonica UNC79MFTsu3.2	excluded	4,374	0	0

"Excluded" genomes have fewer than 10 labelled proteins or problematic labels.

Split. All 486,681 sequences were clustered with MMseqs2 (easy-cluster --min-seq-id 0.40), and whole clusters were assigned at random to 5 folds (seed 0), so proteins sharing ≥40% identity are always in the same fold. Fold sizes: 0: 97,336, 1: 97,339, 2: 97,339, 3: 97,337, 4: 97,330. The released head ProteomeLM-ess was trained with an earlier fold assignment, so evaluate it on the held-out genomes and the minimal cells, not on fold 0.

minimal_cells

The minimal cells of Fig. 5B (NCBI proteomes, locus tags as ids). Labels: E, NE, QE (quasi-essential), uncertain calls (E?, NE?), Not_a_Gene and No_label.

organism	taxid	proteins	E	NE
JCVI-Syn1.0	766747	820	202	386
JCVI-Syn3A	2144189	455	199	64

Sources: Hutchison et al. 2016, Science 351:aad6253, Database S1 (JCVI-Syn1.0) and SynWiki (JCVI-Syn3A).

Usage

import pandas as pd
proteins = pd.read_parquet("hf://datasets/Bitbol-Lab/ProteomeLM-ess-data/data/proteins.parquet")

or datasets.load_dataset("Bitbol-Lab/ProteomeLM-ess-data", "proteins"). To rebuild the input files of the ProteomeLM pipeline (experiments/essentiality in the GitHub repository):

python -m experiments.essentiality.data fetch

Citation

@article{malbranke2026proteomelm,
  title={ProteomeLM: A proteome-scale language model enables accurate and rapid prediction of protein-protein interactions and gene essentiality across taxa},
  author={Malbranke, Cyril and Zalaffi, Gionata Paolo and Bitbol, Anne-Florence},
  journal={Proceedings of the National Academy of Sciences},
  volume={123},
  number={21},
  pages={e2524201123},
  year={2026},
  doi={10.1073/pnas.2524201123}
}

@article{gurumayum2021ogee,
  title={OGEE v3: Online GEne Essentiality database with increased coverage of organisms and human cell lines},
  author={Gurumayum, Sanathoi and Jiang, Puzi and Hao, Xiaowen and Campos, Tulio L and Young, Neil D and Korhonen, Pasi K and Gasser, Robin B and Bork, Peer and Zhao, Xing-Ming and He, Li-jie and Chen, Wei-Hua},
  journal={Nucleic Acids Research},
  volume={49},
  number={D1},
  pages={D998--D1003},
  year={2021},
  doi={10.1093/nar/gkaa884}
}