---
pretty_name: Human Protein Atlas Gene Annotations
license: other
tags:
- biology
- proteins
- human
- human-protein-atlas
- gene-expression
- subcellular-localization
- parquet
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*.parquet
  - split: test
    path: data/test-*.parquet
---

# Human Protein Atlas Gene Annotations

The Human Protein Atlas is an open resource mapping human gene and protein expression across tissues, cells, organs, pathology, and subcellular locations using transcriptomics and antibody-based proteomics.

## Splits

The split is deterministic by Ensembl ID: `sha256(ensembl_id) % 10`. Bucket `0` is `test`; buckets `1` through `9` are `train`.

| Split | Rows |
|---|---:|
| train | 18,138 |
| test | 2,024 |
| total | 20,162 |

## Dataset Statistics

| Field | Value |
|---|---:|
| Gene rows | 20,162 |
| Cancer prognostic metadata rows | 442,504 |
| RNA expression measurement rows | 149,150 |

## Usage

```bash
pip install datasets
```

Load all splits:

```python
from datasets import load_dataset

ds = load_dataset("LiteFold/HumanProteinAtlas")
print(ds)
print(ds["train"][0]["gene"])
```

Load one split:

```python
from datasets import load_dataset

train = load_dataset("LiteFold/HumanProteinAtlas", split="train")
```

Filter genes with protein-level evidence:

```python
from datasets import load_dataset

ds = load_dataset("LiteFold/HumanProteinAtlas", split="train")
protein_level = ds.filter(lambda row: row["evidence"] == "Evidence at protein level")
print(protein_level[0])
```

Load metadata tables directly:

```python
import pandas as pd
from huggingface_hub import hf_hub_download

prognostics_path = hf_hub_download(
    repo_id="LiteFold/HumanProteinAtlas",
    repo_type="dataset",
    filename="metadata/cancer_prognostics.parquet",
)
prognostics = pd.read_parquet(prognostics_path)
print(prognostics.head())

expression_path = hf_hub_download(
    repo_id="LiteFold/HumanProteinAtlas",
    repo_type="dataset",
    filename="metadata/rna_expression_measurements.parquet",
)
expression = pd.read_parquet(expression_path)
print(expression.head())
```

## Key Columns

| Column | Description |
|---|---|
| `ensembl_id` | Ensembl gene ID. |
| `gene` | Gene symbol. |
| `gene_description` | Gene/protein description. |
| `uniprot` | UniProt accession, when available. |
| `chromosome` | Chromosome. |
| `position` | Genomic position string. |
| `evidence` | Main HPA evidence level. |
| `hpa_evidence` | HPA-specific evidence level. |
| `uniprot_evidence` | UniProt evidence level. |
| `nextprot_evidence` | neXtProt evidence level. |
| `antibodies` | HPA antibody IDs. |
| `gene_synonyms` | Gene synonyms. |
| `protein_classes` | Protein class labels. |
| `biological_processes` | Biological process annotations. |
| `molecular_functions` | Molecular function annotations. |
| `disease_involvement` | Disease involvement labels. |
| `subcellular_main_locations` | Main subcellular locations. |
| `subcellular_additional_locations` | Additional subcellular locations. |
| `secretome_locations` | Secretome location annotations. |
| `secretome_functions` | Secretome function annotations. |
| `rna_tissue_specificity` | RNA tissue specificity category. |
| `rna_tissue_distribution` | RNA tissue distribution category. |
| `prognostic_cancer_count` | Number of cancers where the gene is prognostic. |
| `validated_prognostic_cancer_count` | Number of validated prognostic cancer entries. |
| `potential_prognostic_cancer_count` | Number of potential prognostic cancer entries. |
| `prognostic_cancers` | Cancer names with prognostic entries. |
| `split_bucket` | Deterministic split bucket from `sha256(ensembl_id) % 10`. |

# Citation

```
@article{uhlen2015hpa,
  title   = {Tissue-based map of the human proteome},
  author  = {Uhl{\'e}n, Mathias and Fagerberg, Linn and Hallstr{\"o}m, Bj{\"o}rn M. and others},
  journal = {Science},
  volume  = {347},
  number  = {6220},
  pages   = {1260419},
  year    = {2015},
  doi     = {10.1126/science.1260419}
}
```