ConvergeBio/oas-unpaired

OAS Unpaired

The OAS unpaired dataset Observed Antibody Space (OAS), available as parquet with content-defined chunking on HuggingFace.

Configs and Splits

This dataset exposes 91 configs:

Config	Splits	Description
default	heavy, light	All sequences, split by chain
heavy	train	All heavy chain sequences
light	train	All light chain sequences
{Author et al., YYYY}	heavy, light, or both	One author's sequences

from datasets import load_dataset

# All heavy chain sequences
ds = load_dataset("ConvergeBio/oas-unpaired", "heavy", split="train", streaming=True)

# All light chain sequences
ds = load_dataset("ConvergeBio/oas-unpaired", "light", split="train", streaming=True)

# One author
ds = load_dataset("ConvergeBio/oas-unpaired", "Briney et al., 2019", split="heavy")

---

Usage

Load specific columns

from datasets import load_dataset

ds = load_dataset(
    "ConvergeBio/oas-unpaired", "Briney et al., 2019",
    split="heavy",
    columns=["sequence_alignment_aa", "cdr3_aa", "v_call", "j_call", "meta_Subject"],
)
df = ds.to_pandas()

Load a specific author

ds = load_dataset("ConvergeBio/oas-unpaired", "Gidoni et al., 2019", split="heavy")
print(ds)
# Dataset({features: [...], num_rows: ...})

Stream and filter

ds = load_dataset(
    "ConvergeBio/oas-unpaired", "heavy",
    split="train",
    streaming=True,
)

productive = ds.filter(lambda x: x["productive"] == "T")

With DuckDB

import duckdb

con = duckdb.connect()
con.sql("INSTALL httpfs; LOAD httpfs;")
con.sql("CREATE SECRET (TYPE HUGGINGFACE, TOKEN 'hf_...');")

# CDR3 lengths for one author
df = con.sql("""
    SELECT cdr3_aa, length(cdr3_aa) as cdr3_len, v_call, j_call, meta_Subject
    FROM 'hf://datasets/ConvergeBio/oas-unpaired/data/unpaired_heavy/Briney et al., 2019/*.parquet'
    WHERE productive = 'T'
""").df()

# Sequence counts per author across all heavy chains
df = con.sql("""
    SELECT meta_Author, COUNT(*) as n_sequences
    FROM 'hf://datasets/ConvergeBio/oas-unpaired/data/unpaired_heavy/**/*.parquet'
    GROUP BY meta_Author
    ORDER BY n_sequences DESC
""").df()

---

Studies

Each entry is a separate config. Sequence counts come from OAS source file headers. Duplicate rates are based on xxHash128 of sequence_alignment_aa. Click an author name to view the original publication.

Author	Heavy	H Dup%	Light	L Dup%	Summary
Banerjee et al., 2017	3.5M	1.97%	—	—	Rabbit heavy chains from HIV-vaccinated animals; vaccine designed to elicit MPER 4E10/10E8 broadly neutralizing antibodies.
Bashford et al., 2013	258.2K	0.03%	—	—	Human heavy chains from chronic lymphocytic leukemia (CLL) patients; B-cell receptor network analysis via deep sequencing.
Bender et al., 2020	—	—	1.1M	0.43%	Human light chains from bone marrow of POEMS syndrome patients (rare plasma cell disorder with monoclonal immunoglobulin).
Bernardes et al., 2020	3.0M	8.99%	19	—	Human heavy chains from COVID-19 patients; longitudinal multi-omics study of immune cell responses during SARS-CoV-2 infection.
Bernat et al., 2019	2.8M	1.03%	353.3K	0.09%	Human heavy and light chains from healthy PBMC; optimized NGS library preparation for immunoglobulin germline gene inference.
Bhiman et al., 2015	88.5K	1.21%	129.3K	0.02%	Human heavy and light chains from HIV-infected donors tracking viral variants that initiate V1V2-directed broadly neutralizing antibody maturation.
Bolland et al., 2016	27.6K	0.09%	—	—	Mouse pro-B cell heavy chains studying chromatin states that regulate efficient V(D)J recombination at the immunoglobulin locus.
Bonsignori et al., 2016	202.7K	1.6%	—	—	Human heavy chains from HIV-infected donors tracing the germline-to-neutralizer maturation pathway of a CD4-mimicking broadly neutralizing antibody.
Briney et al., 2019	940.3M	7.23%	18	—	Very large human heavy chain dataset (~940M sequences) from 10 healthy donors revealing commonality despite exceptional diversity in baseline human antibody repertoires.
Buchheim et al., 2020	5.2M	0.74%	—	—	Human IgM heavy chains from astronauts and ground controls; studies IgM repertoire plasticity during long-term spaceflight.
Chen et al., 2020	776.4K	6.9%	2.1M	8.11%	Human heavy and light chains from bone marrow of light chain amyloidosis patients; diverse patterns of antibody variable gene disruption.
Collins et al., 2015	359.5K	37.98%	—	—	Mouse heavy chains from spleen showing the murine antibody repertoire is germline-focused and highly variable across individuals.
Corcoran et al., 2016	3.7M	0.16%	1.2M	0.62%	Cross-species (human, mouse C57BL/6, mouse BALB/c, rhesus) heavy and light chains; individualized V gene databases reveal high immunoglobulin gene polymorphism.
Cui et al., 2019	5.5K	11.91%	1.0M	8.76%	Mouse memory B cell heavy and light chains after NP-CGG immunization; models somatic hypermutation targeting in mice.
Davis et al., 2019	11.9M	0.95%	—	—	Human heavy chains from Ebola virus-infected donors across B cell subsets; longitudinal analysis of the B cell response to Ebola infection.
Doria-Rose et al., 2015	526.1K	0.66%	415.3K	0.19%	Human heavy and light chains from HIV-infected donors tracking the developmental pathway of potent V1V2-directed broadly neutralizing antibodies.
Eccles et al., 2020	796	—	13.6K	3.91%	Human heavy and light chains from rhinovirus-reactive (T-bet+) memory B cells linking local cross-reactive IgG to rhinovirus infection.
Eliyahu et al., 2015	1.6M	0.42%	—	—	Human heavy chains from hepatitis C virus (HCV) infected donors identifying immune signatures and potential therapeutic antibody targets.
Ellebedy et al., 2016	11.3M	1.48%	—	—	Human heavy chains from multiple B cell subsets (naive, memory, plasmablasts) defining antigen-specific B cell responses after seasonal influenza vaccination.
Fisher et al., 2017	29.9K	0.27%	130.4K	5.84%	Mouse heavy and light chains from Plasmodium-immunized BALB/c spleen; T-dependent B cell responses forming high-avidity anti-parasite antibodies.
Galson et al., 2015	15.9M	2.72%	—	—	Human heavy chains tracking B cell repertoire dynamics following hepatitis B vaccination.
Galson et al., 2016	16.1M	1.54%	—	—	Human heavy chains studying B cell repertoire dynamics after sequential hepatitis B vaccination; evidence for clonal B cell persistence.
Galson et al., 2016a	4.4M	—	—	—	Human heavy chains from plasma cells after pandemic H1N1 influenza vaccination, investigating AS03 adjuvant effects on B cell repertoire.
Galson et al., 2020	4.6M	2.57%	46	—	Human heavy and light chains from COVID-19 patients; one of the early deep BCR sequencing studies of SARS-CoV-2 infection.
Galson et al., 2015a	3.9M	2.23%	—	—	Human heavy chains across B cell subsets (naive, memory, plasma, unsorted) after meningococcal (MenACWY) conjugate and polysaccharide vaccination.
Ghraichy et al., 2020	8.3M	2.95%	—	—	Human heavy chain repertoire from healthy donors across age groups (children to elderly); studies age-related maturation of immunoglobulin diversity.
Gidoni et al., 2019	13.6M	0.86%	12.7M	28.45%	Human naive B cell heavy and light chains revealing mosaic deletion patterns in the antibody heavy chain gene locus; includes healthy and celiac disease subjects.
Greif et al., 2015	552.0K	1.04%	—	—	Mouse heavy chains across B cell subsets (naive, ASC, plasma cells) after NP-CGG immunization; framework for immune repertoire diversity profiling.
Greiff et al., 2014	3.4M	13.88%	—	—	Mouse plasmablast/plasma cell heavy chains after NP-CGG immunization; quantitative assessment of NGS-based antibody repertoire sequencing robustness.
Greiff et al., 2017	138.8M	5.54%	—	—	Very large mouse heavy chain dataset (multiple strains, antigens, tissues); systems analysis of genetic and antigen-driven predetermination of antibody repertoire structure.
Gupta et al., 2017	3.1M	2.82%	9.3M	6.96%	Human naive B cell heavy and light chains from flu/HepB vaccinated donors; used to evaluate hierarchical clustering methods for identifying B cell clones.
Halliley et al., 2015	593.3K	1.6%	—	—	Human bone marrow plasma cell heavy chains after tetanus/flu vaccination; identifies long-lived plasma cells within the CD19-CD38hiCD138+ subset.
Huang et al., 2016	3.6M	5.57%	3.5M	11.33%	Human memory B cell heavy and light chains from HIV-infected donors; identifies CD4-binding-site antibodies that evolved near-pan HIV neutralization.
Jaffe et al., 2022	1.6M	3.71%	969.5K	37.19%	Human heavy and light chains from COVID-19 and CMV donors; demonstrates that functional antibodies exhibit light chain coherence (light chain pairing bias).
Jiang et al., 2013	3.7M	14.33%	—	—	Human heavy chains from influenza-vaccinated donors; examines lineage structure of the antibody repertoire in response to influenza vaccination.
Johnson et al., 2018	8.6M	2.82%	1.9M	2.83%	Human heavy and light chains from HIV-infected donors; sequences broadly neutralizing antibody exons and introns revealing detailed aspects of antibody evolution.
Joyce et al., 2016	1.6M	—	—	—	Human heavy chains used in a vaccine study identifying broadly protective antibodies that neutralize both group 1 and group 2 influenza A viruses.
Khan et al., 2016	12.0M	17.76%	5	—	Mouse heavy chains from OVA-immunized BALB/c spleen; demonstrates accurate antibody repertoire profiling by molecular amplification fingerprinting.
Kim et al., 2020	45.1M	4.37%	19.6M	10.51%	Large human heavy and light chain dataset from COVID-19 patients; identifies stereotypic VH antibodies that neutralize SARS-CoV-2.
King et al., 2020	13.7M	0.7%	45.9K	14.88%	Human heavy and light chains from tonsillar B cells across subsets (GC, memory, naive, plasmablast); single-cell analysis predicts antibody class switching.
Kuri-Cervantes et al., 2020	8.8M	1.45%	—	—	Human heavy chains from COVID-19 patients; comprehensive mapping of immune perturbations associated with severe COVID-19.
Levin et al., 2016	675.9K	1.49%	—	—	Human heavy chains from allergy patients with/without subcutaneous immunotherapy (SIT); persistence and evolution of allergen-specific IgE repertoires.
Levin et al., 2017	13.1M	5.67%	—	—	Human heavy chains from bone marrow and blood of IgE allergy patients; focuses on bone marrow as an antibody-encoding IgE-producing niche.
Li et al., 2017	1.6M	0.34%	355	0.56%	Heavy and light chain repertoire from Bactrian camels; comparative analysis of conventional and heavy-chain-only (nanobody precursor) antibody repertoires.
Liao et al., 2013	411.6K	1.32%	333.4K	1.63%	Human heavy and light chains from an HIV-infected donor; tracks co-evolution of the broadly neutralizing antibody VRC01 and its founder virus.
Lindner et al., 2015	741.5K	3.7%	—	—	Mouse heavy chains from small intestinal B cells; microbial colonization drives diversification of memory B cells producing secretory IgA.
Meng et al., 2017	32.2M	1.89%	—	—	Human heavy chain atlas across 8 tissues (blood, bone marrow, lung, gut, spleen, etc.); maps B cell clonal distribution throughout the human body.
Menzel et al., 2014	8.3M	22.07%	—	—	Mouse plasmablast/plasma cell heavy chains after NP-CGG immunization; comprehensive evaluation of amplicon library preparation methods for repertoire sequencing.
Montague et al., 2021	10.2M	7.03%	—	—	Human heavy chains from COVID-19 patients; studies dynamics of B cell repertoire and emergence of cross-reactive antibody responses.
Mor et al., 2021	81.1K	7.15%	139.5K	4.37%	Human heavy and light chains from severe COVID-19 patients; identifies multi-clonal SARS-CoV-2 neutralizing antibodies.
Mroczek et al., 2014	121.5K	0.17%	—	—	Human heavy chains across B cell subsets (immature, naive, memory, plasma) from healthy donors; analyzes repertoire composition by B cell subset.
Mukhamedova et al. 2021	447.9K	0.59%	452.4K	5.93%	Human heavy and light chains from RSV prefusion-protein vaccinated donors; studies antibody responses to respiratory syncytial virus (RSV).
Nielsen et al., 2020	12.1M	3.61%	—	—	Human heavy chains from COVID-19 patients and nasopharyngeal swabs; studies clonal B cell expansion and convergent antibody responses to SARS-CoV-2.
Ohm-Laursen et al., 2018	7.2M	5.43%	—	—	Human heavy chains from bronchial biopsies and blood of asthma patients; studies local clonal B cell diversification and dissemination in the airway.
Ota et al., 2010	—	—	20.1K	9.88%	Mouse light chains from healthy spleen; studies how BAFF regulates B cell receptor repertoire composition and self-reactivity.
Palanichamy et al., 2014	339.6K	0.12%	—	—	Human heavy chains from cerebrospinal fluid and blood of multiple sclerosis patients; immunoglobulin class-switched B cells form a CNS-periphery immune axis.
Parameswaran et al., 2014	314.1K	4.52%	—	—	Human heavy chains from dengue fever and non-dengue febrile illness donors; identifies convergent antibody signatures across multiple individuals.
Prohaska et al., 2018	255.0K	1.37%	—	—	Mouse heavy chains from B cell subsets (B-1a, B-1b, B-2, follicular, marginal zone) in peritoneal cavity and spleen; highlights innate-like B cell repertoire differences.
Rettig et al., 2018	27.8K	2.67%	30.8K	17.56%	Mouse heavy and light chains from healthy spleen; naive repertoire characterization using unamplified (no PCR) high-throughput sequencing to minimize amplification bias.
Richardson et al., 2022	406.9K	3.16%	—	—	Heavy chains from Kymouse (humanized transgenic) naive splenic B cells; characterizes the human-like immune repertoire in this model organism.
Rubelt et al., 2016	2.2M	0.87%	—	—	Human heavy chains from memory and naive B cells in twins; heritable individual differences drive unique B cell receptor repertoire formation.
Schanz et al., 2014	4.3M	3.36%	1.7M	0.84%	Human heavy and light chains from HIV-infected donors using isotype-specific (IgG, IgM) high-throughput immunoglobulin sequencing.
Schultheiss et al., 2020	4.7M	0.21%	—	—	Human heavy chains from COVID-19 patients; next-generation sequencing of both T and B cell receptor repertoires from COVID-19 patients and healthy controls.
Setliff et al., 2018	22.5M	2.91%	1.9M	0.8%	Large longitudinal human heavy and light chain dataset from HIV-infected donors; reveals stable clonal memory B cell pools across multiple donors.
Sevy et al., 2019	18.7M	0.4%	74	1.35%	Human heavy chains from HIV-infected and flu-vaccinated donors; repertoire fingerprinting by PCA reveals shared clonotypes across individuals.
Sheng et al., 2017	541.8K	2.6%	755.9K	13.83%	Human heavy and light chains from healthy PBMC; describes gene-specific amino acid substitution profiles quantifying somatic hypermutation type and frequency.
Simonich et al., 2020	847.0K	0.05%	1.2M	1.42%	Human heavy and light chains from HIV-infected infants; kappa light chain maturation drives rapid development of broadly neutralizing antibodies.
Soto et al., 2016	333.9K	2.99%	422.3K	—	Human heavy and light chains from HIV-infected donors; traces the developmental pathway of the MPER-directed broadly neutralizing antibody 10E8.
Soto et al., 2019	553.3M	19.91%	242.7M	25.32%	Very large human heavy and light chain dataset (~796M sequences) from healthy donors; demonstrates high frequency of shared clonotypes in human B cell receptor repertoires.
Stern et al., 2014	10.1M	8.62%	207	0.48%	Human heavy chains from multiple sclerosis patient brain lesions and draining cervical lymph nodes; B cells populating the MS brain mature in the CNS.
Sundling et al., 2014	130.2K	0.23%	—	—	Rhesus macaque IgG-switched heavy chains from PBMC after HIV vaccination; single-cell and deep sequencing reveals diverse antibody responses.
Tipton et al., 2015	15.7M	1.58%	—	—	Human heavy chains from SLE patients and healthy controls; studies diversity, cellular origin, and autoreactivity of antibody-secreting cells.
Tong et al., 2017	59.3K	0.21%	—	—	Mouse heavy chains from OVA-immunized spleen and bone marrow; studies how IgH isotype-specific B cell receptor expression influences B cell fate.
Turchaninova et al., 2015	201.4K	0.01%	—	—	Human heavy chains from memory, naive, and plasma B cells; demonstrates high-quality full-length immunoglobulin profiling using unique molecular barcodes.
Turner et al., 2021	1.6M	0.64%	11.7K	4.66%	Human heavy and light chains from germinal center B cells and plasmablasts after SARS-CoV-2 mRNA vaccination; persistent germinal center responses observed.
VanDuijn et al., 2017	5.2M	1.5%	9	—	Rat heavy chains from DNP/HuD-immunized spleen; studies immune repertoire by combining next-generation sequencing with protein mass spectrometry.
Vander Heiden et al., 2017	2.5M	3.36%	5.3M	9.48%	Human heavy and light chains from myasthenia gravis (AChR-MG and MuSK-MG) patients; B cell repertoire dysregulation in autoimmune disease.
Vergani et al., 2017	13.5M	5.94%	—	—	Human heavy chains from healthy naive and unsorted B cells; presents a novel high-throughput method for full-length IGHV-D-J sequencing.
Waltari et al., 2018	29.6M	0.78%	45.1M	6.55%	Large heavy and light chain dataset from HIV-infected donors and humanized mice across multiple tissues; 5' RACE amplification maps B cell receptor features.
Wesemann et al., 2013	37.0K	1.9%	29.9K	27.03%	Mouse heavy and light chains from gut lamina propria, bone marrow, and spleen; gut microbial colonization influences early B cell lineage development.
Woodruff et al., 2020	18.3K	4.11%	45.5K	1.97%	Human heavy and light chains from antibody-secreting and naive B cells in COVID-19; extrafollicular B cell responses correlate with neutralizing antibodies and morbidity.
Wu et al., 2011	271.5K	2.83%	37.8K	—	Human heavy and light chains from HIV-infected donors; focused evolution of broadly neutralizing antibodies revealed by structures and deep sequencing.
Wu et al., 2014	37.5K	2.07%	—	—	Human heavy chains from allergic rhinitis patients in- and out-of-season; seasonal grass pollen exposure shapes local and peripheral blood IgE repertoires.
Wu et al., 2015	1.4M	15.47%	827.2K	5.74%	Human heavy and light chains from a single HIV-infected donor over 15 years; tracks maturation and diversification of the VRC01 broadly neutralizing antibody lineage.
Zhou et al., 2013	302.8K	1.14%	691.0K	1.19%	Human heavy and light chains from multiple HIV-infected donors; multidonor analysis of structural elements, genetic determinants, and maturation of broadly neutralizing antibodies.
Zhou et al., 2015	383.2K	1.01%	—	—	Human heavy chains from HIV-infected donors; structural repertoire of antibodies targeting the CD4 supersite on HIV-1.
Zhu et al., 2012	200.0K	0.51%	115.1K	—	Human heavy and light chains from HIV-infected donors; identifies somatic populations of PGT135-137 broadly neutralizing antibodies by deep sequencing.
Zhu et al., 2013	533.7K	5.1%	478.8K	9.19%	Human heavy and light chains from HIV-infected donors; de novo identification of VRC01-class HIV-1 broadly neutralizing antibodies by next-generation sequencing.

Total: 2,070,782,127 heavy + 356,864,753 light sequences across all studies.

---

Schema

Each row is one antibody sequence. Fields follow the AIRR Community standard, with OAS study metadata.

Core AIRR fields

Column	Type	Description
sequence	string	Raw nucleotide sequence
locus	string	IGH, IGK, or IGL
v_call, d_call, j_call	string	V/D/J gene assignments
sequence_alignment	string	Aligned nucleotide sequence
sequence_alignment_aa	string	Aligned amino acid sequence
junction	string	Junction nucleotides
junction_aa	string	Junction amino acids
cdr1_aa, cdr2_aa, cdr3_aa	string	CDR amino acid sequences
fwr1_aa … fwr4_aa	string	Framework amino acid sequences
v_identity, d_identity, j_identity	double	Alignment identity scores
productive	string	Whether the sequence is productive
stop_codon, vj_in_frame, v_frameshift	string	QC flags
Redundancy	int64	Copy count in original OAS study
ANARCI_numbering	string	ANARCI antibody numbering
ANARCI_status	string	ANARCI annotation status

OAS metadata columns (meta_*)

Column	Description
meta_Run	SRA run accession
meta_Author	Author label (matches config name)
meta_Species	Donor species
meta_Age	Donor age
meta_BSource	B-cell source tissue
meta_BType	B-cell type
meta_Vaccine	Vaccine/antigen if applicable
meta_Disease	Disease condition
meta_Subject	Subject identifier
meta_Longitudinal	Whether study is longitudinal
meta_Isotype	Isotype
meta_Chain	Heavy or Light
meta_Link	URL to original OAS study page

Hash columns

Column	Type	Description
aa_hash_hi	uint64	High 64 bits of xxh128(sequence_alignment_aa)
aa_hash_lo	uint64	Low 64 bits of xxh128(sequence_alignment_aa)

---

Citation

If you use this dataset, please cite the original OAS publication:

@article{Olsen2022,
  author  = {Olsen, Tobias H. and Boyles, Fergus and Deane, Charlotte M.},
  title   = {Observed Antibody Space: A diverse database of cleaned, annotated, and translated unpaired and paired antibody sequences},
  journal = {Protein Science},
  year    = {2022},
  volume  = {31},
  number  = {1},
  pages   = {141--146},
  doi     = {10.1002/pro.4205}
}

Please also cite the individual studies whose data you use -- links are available in the meta_Link column and on the OAS website.

About

Built by Converge Bio — accelerating drug discovery with generative AI. Converge Bio develops foundation models for protein engineering, antibody design, and gene expression optimization, powering its computational lab products ConvergeAB, ConvergeGEO, and ConvergeCELL.

License

OAS data is available under CC BY 4.0.