# Enformer Model, GTEx Fine-Mapped eQTLs and 1000 Genomes Variant Effect Scores

## Facts

- Provider: DeepMind and Calico Life Sciences
- License: CC-BY-4.0 (https://storage.googleapis.com/dm-enformer/README)
- Format: TSV (in sampled files)
- Stored at: Google Cloud Storage, gs://dm-enformer (us)
- Access: public, readable without an account
- DOI: https://doi.org/10.1038/s41592-021-01252-x
- Category: Genomics & Bioinformatics
- Source: https://doi.org/10.1038/s41592-021-01252-x
- Page: https://scigantic.com/archive/0577c367-38fb-5c8b-bbd9-0b10b4291dbc
- Keywords: Genomics & Bioinformatics, TSV, Google Cloud Storage

## About

Public release accompanying Enformer (Avsec et al., Nature Methods 2021), DeepMind and Calico's transformer model that predicts gene expression and chromatin tracks from up to about 100 kb of DNA sequence context by integrating long-range interactions. Measured 2026-11-13, the bucket has three top-level folders. models/enformer/ holds the TensorFlow SavedModel (saved_model.pb, variables/) and Sonnet weights, plus fitted robust-scaler and PCA transforms for SAD (SNP activity difference) scores. data/gtex_fine/ holds GTEx fine-mapped eQTL sets per tissue: SuSiE output tables (.tsv, tens of MB per tissue) and positive/negative variant VCFs per tissue used to evaluate variant effect prediction. variant-scores/1000-genomes/enformer/ holds precomputed Enformer variant effect predictions for 1000 Genomes variants with minor allele frequency above 0.005, as one HDF5 file per chromosome (individual files range up to about 42 GB), plus a PC20 subfolder. Open the HDF5 files with h5py, the TSVs with pandas, the VCFs with pysam or cyvcf2, and the model with tensorflow / tensorflow_hub. Use cases: variant effect prediction, eQTL prioritization, regulatory genomics and sequence-to-expression deep learning. Copyright 2021 DeepMind Technologies, released under CC BY 4.0 per the in-bucket README; cite the Enformer paper.

## Load the data

```python
import os
import pandas as pd
df = pd.read_csv(os.path.join(os.environ['SCIGANTIC_MOUNT_PATH'], 'data/gtex_fine/susie/Adipose_Subcutaneous.tsv'), nrows=1000)
df.head()
```

Run in a Scigantic notebook on 2026-11-14 against `data/gtex_fine/susie/Adipose_Subcutaneous.tsv`: CSV/TSV: columns ['Unnamed: 0', 'chromosome', 'start', 'end', 'variant', 'allele1', 'allele2', 'minor_allele', 'tissue', 'gene', 'maf', 'beta_marginal'], dtypes ['int64', 'object', 'int64', 'int64', 'object', 'object', 'object', 'object', 'object', 'object', 'float64', 'float64'], first 1000 rows re.

From any machine:

```python
import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('dm-enformer/', detail=False)[:10]
```

Lists the files with anonymous access. Runs on any machine, with no cloud account.

## Data structure

File types in sampled files: TSV

Example file: `data/gtex_fine/susie/Adipose_Subcutaneous.tsv`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/0577c367-38fb-5c8b-bbd9-0b10b4291dbc
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/0577c367-38fb-5c8b-bbd9-0b10b4291dbc
- Over MCP: `get_schema_card` with id `0577c367-38fb-5c8b-bbd9-0b10b4291dbc` at https://api.scigantic.com/mcp

## Related topics

- [Genomics & Bioinformatics datasets](https://scigantic.com/topics/genomics-and-bioinformatics)
- [TSV datasets](https://scigantic.com/topics/tsv)
