Resource

CD4 TCR atlas

Published alpha/beta T cell receptors of CD4 T cells, human and mouse, gathered in one table with what is known about what each receptor recognises: the presenting molecule, the antigen or epitope, and how the specificity was established.

Search the annotated receptors

The search runs in your browser on the receptors that carry an antigen or an epitope. Nothing you type is sent to a server. The unannotated repertoires are in the download files only.

What the annotation levels mean

• Epitope: a defined peptide (or lipid) is mapped for that receptor, usually with the presenting molecule.
• Antigen: the receptor comes from cells sorted with a peptide:MHC tetramer, from cells that expressed activation markers after stimulation with an antigen or a peptide pool, or from a clone that responds to an extract. Activation-based sorts include bystander cells.
• Condition: only a disease label is known.
• None: a CD4 repertoire without antigen information (sorted subsets, tissue or blood CD4 T cells).

The column specificity_evidence states how specificity was established for each record, and cd4_evidence why the receptor is counted as CD4. Read them before drawing conclusions: a reporter assay, a tetramer and an activation marker are not the same evidence.

Sources

Every record keeps the name of its source and its reference. Please cite the original studies and databases when you use their receptors.

Download

Tables are split so that each file stays under 25 MB. The parquet parts of one table have the same columns and can be read together, for example with pandas.read_parquet on each part or with DuckDB: select * from 'cd4_tcr_atlas_records-part*.parquet'.

Columns of the records table:

The scripts that build the atlas from the original files, one parser per study, are in code/build_atlas.py and code/parsers/ (each parser states the paper, what is loaded, what is left out and why). The original files are not redistributed here: get them from the sources.

Limits

• Most records carry no antigen: one study of sorted helper subsets from healthy donors (Kasatskaya 2020) is about 70% of the table.
• VDJdb, IEDB and McPAS-TCR overlap. The clonotype table merges identical sequences; the records table does not.
• Donor labels are local to each source and are never merged across sources. Some are a sequencing run or a pool of samples.
• Two old studies (1993, 2002) use the V and J gene names of their time; their junctions were transcribed from printed figures and tables.
• CDR3 are stored as published. The columns cdr3a_core and cdr3b_core hold the CDR3 without the conserved C and F/W anchors, for matching across conventions.
• Datasets under controlled access (dbGaP, EGA, immuneACCESS) are not included, and several public datasets were left out because CD4 T cells cannot be told apart in the deposited files.

Rights. This is a compilation of receptor sequences and annotations reported in public databases and publications. Each source keeps its own terms of use; check them before reuse, in particular for commercial use. If you are the author of a dataset and want its records corrected or removed, the source name and reference of every record make this straightforward: write to kilian.maire@sund.ku.dk.