Protein Annotation · LiteFold
HumanProteinAtlas
Tabular gene and protein expression annotations across human tissues, pathology, and subcellular compartments.
What HumanProteinAtlas is
The Human Protein Atlas dataset provides a comprehensive mapping of human gene and protein expression across cells, tissues, organs, pathology, and subcellular compartments. Packaged by LiteFold in tabular format, it compiles annotations derived from transcriptomics and antibody-based proteomics for 20,162 human genes.
This resource addresses the challenge of accessing integrated expression, localisation, and pathology data for human proteins in standardised machine-readable tables. It allows researchers in biology, medicine, and genomics to analyse protein distributions, functional classes, and disease associations without custom parsing of raw atlas releases.
How it was made
The main dataset is organised into tabular gene records containing identifiers, genomic positions, evidence levels from the Human Protein Atlas, UniProt, and neXtProt, as well as functional and localisation tags. Samples are deterministically partitioned into training and testing sets based on the SHA-256 hash of each Ensembl gene identifier modulo 10, reserving bucket 0 for testing and buckets 1 to 9 for training.
In addition to the primary gene annotation table of 20,162 rows, the collection includes separate Parquet metadata tables covering 442,504 cancer prognostic records and 149,150 RNA expression measurement entries.
Key facts
- Publisher
- LiteFold
- Total gene records
- 20,162
- Train split size
- 18,138
- Test split size
- 2,024
- Cancer prognostic rows
- 442,504
- RNA expression rows
- 149,150
- Primary format
- Parquet
- Modality
- Tabular, text
What you can use it for
- Subcellular Localisation Analysis
- Query main and additional subcellular compartments, secretome locations, and secretome functions across catalogued human genes.
- Cancer Prognostic Profiling
- Examine associations between gene expression and cancer outcomes using validated and potential prognostic counts across specific cancer types.
- Tissue Specificity Classification
- Filter and evaluate genes according to RNA tissue specificity and tissue distribution categories.
- Protein Evidence Filtering
- Identify genes with validated experimental support by filtering on HPA, UniProt, or neXtProt protein-level evidence strings.
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/HumanProteinAtlas at the moment you asked. Nothing is cached or stored — every row above came from that request.
Strengths and limits
Good at
- Provides deterministic, hash-based train and test splits on Ensembl gene identifiers for reproducible machine learning workflows.
- Combines protein-level evidence from multiple resources including the Human Protein Atlas, UniProt, and neXtProt.
- Includes extensive supplementary metadata tables for cancer prognostics and detailed RNA expression measurements.
Watch out for
- UniProt accession codes are not available for every gene record.
- The dataset focuses exclusively on human genes and proteins, precluding comparative cross-species analyses without external data.
- Cancer prognostic and RNA expression measurements are stored in separate metadata tables rather than within the primary gene split rows.
Questions
Can I look inside HumanProteinAtlas 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 HumanProteinAtlas under?
Its authors declare other. Read the terms on the original page before reusing the data.
What evidence levels are included for protein annotations?
The dataset provides distinct columns for the main Human Protein Atlas evidence level, HPA-specific evidence, UniProt evidence, and neXtProt evidence.
How are the training and testing sets split?
The split is assigned deterministically using the formula sha256(ensembl_id) % 10, with bucket 0 designated as the test set and buckets 1 through 9 designated as the training set.
What formats and tables are available?
The dataset provides a primary gene table in Parquet format, accompanied by separate Parquet tables containing 442,504 cancer prognostic records and 149,150 RNA expression measurement rows.
How to cite it
@article{uhlen2015hpa, title = {Tissue-based map of the human proteome}, author = {Uhlén, Mathias and Fagerberg, Linn and Hallström, Björn M. and others}, journal = {Science}, volume = {347}, number = {6220}, pages = {1260419}, year = {2015}, doi = {10.1126/science.1260419}}
Written from HumanProteinAtlas'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.