Immunology · LiteFold
IEDB
A tabular export of experimentally characterised immune epitopes and assays across diverse immunological contexts.
What IEDB is
This dataset provides a structured export of experimentally characterised immune epitopes derived from the Immune Epitope Database (IEDB), published by LiteFold. It compiles assay-level experimental observations covering B-cell, T-cell, and major histocompatibility complex (MHC) binding interactions across infectious diseases, allergies, autoimmune conditions, and transplantation.
The resource converts uncompressed XML database dumps into a standardised tabular Parquet format. Spanning millions of assay measurements across varied hosts and organisms, it enables computational immunologists and machine learning practitioners to train, validate, and benchmark models for epitope recognition and antigen presentation.
How it was made
The dataset is compiled by parsing 26,785 XML files into 7,450,811 assay records. Each row captures reference metadata, epitope identifiers, structural and chemical classifications, linear or discontinuous sequence details, host and disease background, MHC allele annotations, and experimental outcomes reported both qualitatively and quantitatively.
Partitioning is deterministic and epitope-aware: a SHA-256 hash of the epitope identifier modulo 10 routes bucket 0 to the test split (745,344 rows) and buckets 1 through 9 to the training split (6,705,467 rows). If an epitope identifier is absent, the hashing falls back sequentially to the reference identifier, assay identifier, or XML filename.
Key facts
- Publisher
- LiteFold
- Total rows
- 7,450,811
- Unique epitopes
- 2,320,500
- Unique references
- 26,785
- Data format
- Parquet
- Training split rows
- 6,705,467
- Test split rows
- 745,344
- Largest assay category
- MhcLigandElution (4,635,502 rows)
- Predominant chemical type
- Peptide from protein (6,773,619 rows)
What you can use it for
- T-cell epitope discovery
- Evaluate algorithms that predict cellular immune responses using 570,391 curated T-cell assay results coupled with specific MHC allele and host annotations.
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/IEDB at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 6,705,467 rows of default/train
antigen_evidence_code
string_label- Exact match to reference information25%
- Representative selection1%
- Not determined0%
73.3% empty
article_title
string_text- min
- 23
- median
- 108
- mean
- 109.83
- max
- 347
22.4% empty
article_year
int- min
- 1,952
- median
- 2,020
- mean
- 2018.74
- max
- 2,026
22.4% empty
assay_category
string_label- MhcLigandElution62%
- BCell19%
- MhcBinding11%
assay_comments
string_labelNo distribution reported for this column.
100.0% empty
assay_id
string_text- min
- 2
- median
- 8
- mean
- 7.61
- max
- 8
assay_location_of_data
string_text- min
- 4
- median
- 15
- mean
- 15.82
- max
- 126
18.3% empty
assay_type_id
string_label- 9153%
- 55012%
- 5729%
authors
string_text- min
- 5
- median
- 183
- mean
- 200.57
- max
- 2,347
22.4% empty
cell_culture_conditions
string_label- Direct Ex Vivo5%
- Short Term Restimulated2%
- Cell Line / Clone1%
92.5% empty
cell_tissue_type
string_label- blood5%
- spleen1%
- lymph node0%
92.5% empty
cell_type
string_label- PBMC2%
- T cell CD4+2%
- T cell CD8+1%
92.5% empty
date_last_updated
string_label- 2026-02-27-08:0013%
- 2026-02-26-08:009%
- 2025-11-06-08:009%
12.8% empty
discontinuous_residues
string_text- min
- 2
- median
- 82
- mean
- 76.81
- max
- 419
99.2% empty
disease_state
string_label- ONTIE:00034235%
- DOID:00806004%
- DOID:29453%
81.2% empty
ending_position
int- min
- 3
- median
- 258
- mean
- 595.44
- max
- 34,833
12.0% empty
epitope_chemical_type
string_label- Peptide from protein91%
- Peptide, no natural source8%
- Discontinuous protein residues1%
epitope_evidence_code
string_label- Representative selection16%
- Author provided Identifier8%
- Not determined3%
72.9% empty
epitope_id
string_text- min
- 1
- median
- 6
- mean
- 6.37
- max
- 7
0.00% empty
epitope_name
string_text- min
- 1
- median
- 12
- mean
- 14.46
- max
- 85
20.2% empty
epitope_structure_defines
string_label- Exact Epitope72%
- Epitope containing region/antigenic site27%
- Partial Epitope1%
0.02% empty
epitope_structure_type
string_label- FragmentOfANaturalSequenceMolecule91%
- SequenceMoleculeNoNaturalSource8%
- DiscontinuousRegionOnAccessionSequenceMolecule1%
host_age
string_text- min
- 1
- median
- 11
- mean
- 10.62
- max
- 39
91.4% empty
host_organism_id
string_label- 960622%
- 100000001%
- 100000671%
73.3% empty
host_sex
string_label- F2%
- M1%
97.6% empty
immunogen_evidence_code
string_label- Representative selection19%
- Exact match to reference information3%
- Not determined1%
76.3% empty
is_positive
bool- True74%
- False26%
journal_title
string_text- min
- 3
- median
- 23
- mean
- 25.37
- max
- 197
22.4% empty
linear_sequence
string_text- min
- 1
- median
- 12
- mean
- 12.21
- max
- 829
1.1% empty
linear_sequence_length
int- min
- 1
- median
- 12
- mean
- 12.21
- max
- 829
1.1% empty
mhc_allele_id
string_label- 30324%
- 30411%
- 1433%
31.9% empty
mhc_allele_types_present
string_text- min
- 1
- median
- 13
- mean
- 14.42
- max
- 150
67.4% empty
pubmed_id
string_text- min
- 5
- median
- 8
- mean
- 7.98
- max
- 8
22.4% empty
qualitative_measurement
string_label- Positive70%
- Negative26%
- Positive-Low2%
quantitative_measurement
string_text- min
- 3
- median
- 6
- mean
- 5.61
- max
- 25
96.0% empty
reference_id
string_text- min
- 1
- median
- 7
- mean
- 6.99
- max
- 7
reference_region
string_text- min
- 2
- median
- 87
- mean
- 80.81
- max
- 409
99.4% empty
source_molecule_genbank_id
string_text- min
- 4
- median
- 8
- mean
- 8.64
- max
- 14
11.2% empty
source_organism_id
string_text- min
- 1
- median
- 4
- mean
- 4.61
- max
- 8
8.2% empty
split_bucket
int- min
- 1
- median
- 5
- mean
- 5.00
- max
- 9
starting_position
int- min
- 0
- median
- 247
- mean
- 584.30
- max
- 34,825
12.0% empty
xml_file
string_text- min
- 5
- median
- 11
- mean
- 10.99
- max
- 11
Strengths and limits
Good at
- Uses an epitope-aware deterministic split to help prevent evaluation data leakage between training and testing sets.
- Maintains extensive experimental context per row, including host species, disease state, MHC alleles, cell culture conditions, and curation evidence codes.
Watch out for
- Peptide epitopes derived from natural proteins dominate the dataset, leaving non-peptidic antigens such as carbohydrates represented in much smaller numbers.
- Contextual fields such as host age, sex, and quantitative measurement values are only populated when explicitly available in the underlying curated literature.
Questions
Can I look inside IEDB 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 IEDB under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
What assay types are contained in the dataset?
The records are divided into four assay categories: MHC ligand elution (4,635,502 rows), B-cell assays (1,419,468 rows), MHC binding assays (825,450 rows), and T-cell assays (570,391 rows).
How are the training and test sets partitioned?
Splits are assigned deterministically using `sha256(epitope_id) % 10`. Bucket 0 forms the test split, while buckets 1 to 9 form the training split, cascading to reference ID, assay ID, or XML filename if an epitope ID is missing.
Does the export contain discontinuous or non-peptidic epitopes?
Yes. Although linear peptides from proteins form the vast majority, the dataset includes 51,083 discontinuous protein residue entries, 14,279 other non-sequence structures, and 5,650 carbohydrate fragments.
How can very large subsets of the data be processed efficiently?
The dataset can be streamed row-by-row or queried directly in Parquet format using columnar analysis tools such as DuckDB, PyArrow, Polars, or Spark.
How to cite it
Vita, Randi and others. The Immune Epitope Database (IEDB): 2024 update. Nucleic Acids Research, 53(D1), D436–D443, 2025. doi: 10.1093/nar/gkae1092.
Written from IEDB'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.