README.md

---
dataset_info:
- config_name: ancestry_prediction
  features:
  - name: sequence
    dtype: string
  - name: id
    dtype: string
  - name: start_idx
    dtype: int64
  - name: chromosome
    dtype: string
  - name: variants
    dtype: string
  - name: parents
    dtype: string
  - name: label
    dtype: int64
  splits:
  - name: train
    num_bytes: 914858176
    num_examples: 14085
  download_size: 298355655
  dataset_size: 914858176
- config_name: coding_pathogenicity
  features:
  - name: alt_left
    dtype: string
  - name: alt_right
    dtype: string
  - name: ref_left
    dtype: string
  - name: ref_right
    dtype: string
  - name: label
    dtype: float64
  - name: chrom
    dtype: string
  splits:
  - name: train
    num_bytes: 353452168
    num_examples: 82872
  download_size: 128733817
  dataset_size: 353452168
- config_name: common_vs_rare
  features:
  - name: ref_forward_sequence
    dtype: string
  - name: alt_forward_sequence
    dtype: string
  - name: label
    dtype: int8
  - name: chromosome
    dtype: string
  - name: position
    dtype: int32
  - name: alt_left
    dtype: string
  - name: alt_right
    dtype: string
  - name: ref_left
    dtype: string
  - name: ref_right
    dtype: string
  - name: chrom
    dtype: string
  splits:
  - name: train
    num_bytes: 1239819384
    num_examples: 200000
  download_size: 542989051
  dataset_size: 1239819384
- config_name: expression
  features:
  - name: alt_left
    dtype: string
  - name: alt_right
    dtype: string
  - name: ref_left
    dtype: string
  - name: ref_right
    dtype: string
  - name: label
    dtype: int64
  - name: chrom
    dtype: string
  - name: deepsea_label
    dtype: string
  splits:
  - name: train
    num_bytes: 650249506
    num_examples: 156162
  download_size: 270672126
  dataset_size: 650249506
- config_name: meqtl
  features:
  - name: label
    dtype: int8
  - name: chromosome
    dtype: string
  - name: position
    dtype: int32
  - name: uid
    dtype: string
  - name: alt_left
    dtype: string
  - name: alt_right
    dtype: string
  - name: ref_left
    dtype: string
  - name: ref_right
    dtype: string
  - name: chrom
    dtype: string
  splits:
  - name: train
    num_bytes: 318695785
    num_examples: 76488
  download_size: 134361097
  dataset_size: 318695785
- config_name: non_coding_pathogenicity
  features:
  - name: alt_left
    dtype: string
  - name: alt_right
    dtype: string
  - name: ref_left
    dtype: string
  - name: ref_right
    dtype: string
  - name: label
    dtype: int64
  - name: chrom
    dtype: string
  - name: vep
    dtype: string
  splits:
  - name: train
    num_bytes: 1265476277
    num_examples: 295495
  download_size: 442285117
  dataset_size: 1265476277
- config_name: sqtl
  features:
  - name: geneId
    dtype: string
  - name: snpId
    dtype: string
  - name: pv
    dtype: float64
  - name: label
    dtype: int64
  - name: tissue
    dtype: string
  - name: __index_level_0__
    dtype: int64
  - name: chrom
    dtype: string
  - name: pos
    dtype: int64
  - name: ref_allele
    dtype: string
  - name: alt_allele
    dtype: string
  - name: alt_left
    dtype: string
  - name: alt_right
    dtype: string
  - name: ref_left
    dtype: string
  - name: ref_right
    dtype: string
  splits:
  - name: train
    num_bytes: 4501372944
    num_examples: 1055241
  download_size: 1855898562
  dataset_size: 4501372944
configs:
- config_name: ancestry_prediction
  data_files:
  - split: train
    path: ancestry_prediction/train-*
- config_name: coding_pathogenicity
  data_files:
  - split: train
    path: coding_pathogenicity/train-*
- config_name: common_vs_rare
  data_files:
  - split: train
    path: common_vs_rare/train-*
- config_name: expression
  data_files:
  - split: train
    path: expression/train-*
- config_name: meqtl
  data_files:
  - split: train
    path: meqtl/train-*
- config_name: non_coding_pathogenicity
  data_files:
  - split: train
    path: non_coding_pathogenicity/train-*
- config_name: sqtl
  data_files:
  - split: train
    path: sqtl/train-*
tags:
- gfm
- genomics
- dna
- genomic foundation model
- m42
- BioFM
- BioToken
- Variant
license: cc-by-nc-4.0
task_categories:
- text-classification
pretty_name: variant-benchmark
size_categories:
- 1M<n<10M
---

# Variant Benchmark
This benchmark is designed to evaluate how effectively models leverage variant information across diverse biological contexts. 
Unlike conventional genomic benchmarks that focus primarily on region classification, our approach extends to a broader range of variant-driven molecular processes.


Existing assessments, such as [BEND]((https://github.com/frederikkemarin/BEND)) and the [Genomic Long-Range Benchmark (GLRB)](https://huggingface.co/datasets/InstaDeepAI/genomics-long-range-benchmark), 
provide valuable insights into specific tasks like noncoding pathogenicity and tissue-specific expression. 
However, they do not fully capture the complexity of variant-mediated effects across multiple biological mechanisms. 
This benchmark addresses that gap by incorporating a more comprehensive set of evaluations, enabling a deeper assessment of functional genomics models.


## Quick Start

```python
from datasets import load_dataset

DATASET_SUBSET = "ancestry_prediction"

# Dataset subset should be from one of the available tasks:
# ['ancestry_prediction', 'non_coding_pathogenicity', 'expression',
#                         'common_vs_rare', 'coding_pathogenicity', 'meqtl', 'sqtl']

ds = load_dataset(
      "m42-health/variant-benchmark",
      DATASET_SUBSET,
)

print(ds)

# Example output:
# (Each dataset has a default 'train' split, but we recommend using k-fold cross-validation for better evaluation)
# DatasetDict({
#     train: Dataset({
#         features: ['sequence', 'id', 'start_idx', 'chromosome', 'variants', 'parents', 'label'],
#         num_rows: 14085
#     })
# })

```

## Benchmark Tasks

- **Coding pathogenicity assessment:** `subset: coding_pathogenicity`<br/>
Accurate prediction of pathogenic coding variants is fundamental to precision medicine and clinical genomics. 
For this task, we use the [AlphaMissense](https://github.com/google-deepmind/alphamissense) dataset, which provides a comprehensive catalog of coding variants annotated for pathogenicity.

- **Noncoding pathogenicity assessment:** `subset: non_coding_pathogenicity`<br/>
Pathogenic variants in noncoding regions significantly impact gene regulation, influencing many complex traits and diseases. 
We assess this using the [BEND dataset](https://github.com/frederikkemarin/BEND), which contains 295,000 annotated single-nucleotide variants (SNVs) in noncoding genomic regions.

- **Expression effect prediction:** `subset: expression`<br/>
Variant-driven changes in gene expression contribute to phenotypic diversity and disease processes. 
To quantify these effects, we use gene expression data from [DeepSea](https://www.nature.com/articles/nmeth.3547), which provides variant-associated regulatory annotations.

- **Alternative splicing:** `subset: sqtl`<br/>
Variant-induced alternative splicing contributes significantly to human proteomic diversity and affects biological processes and diseases. 
We evaluate splicing-related variant effects using an [sQTL dataset](https://www.nature.com/articles/s41467-020-20578-2) derived from sqtlSeeker2, containing over one million variant-tissue pairs.

- **DNA methylation:** `subset: meqtl`<br/>
Genomic variants can influence DNA methylation patterns, affecting gene regulation and disease susceptibility.
For this task, we utilize meQTL data from the [GRASP database](https://pubmed.ncbi.nlm.nih.gov/24931982/), which links genetic variants to methylation changes.

- **Ancestry classification:** `subset: ancestry_prediction`<br/>
Genomic variation encodes population structure, informing studies in evolutionary biology and disease susceptibility.
To evaluate this capability, we used genomic segments labeled by five major superpopulations from the 1000 Genomes Project.

- **Common vs synthetic variants:** `subset: common_vs_rare`<br/>
This task evaluates the model’s ability to recognize biologically conserved genomic contexts characteristic of authentic common variants. 
To create this deataset, we randomly sampled 100K common variants (MAF > 0.05) from GnomAD~\citep{chen2024gnomad} and paired each with a synthetic control variant generated by randomly substituting a nucleotide within a ±20-nucleotide local context window.


## Citation
```
@article {Medvedev2025.03.27.645711,
    author = {Medvedev, Aleksandr and Viswanathan, Karthik and Kanithi, Praveenkumar and Vishniakov, Kirill and Munjal, Prateek and Christophe, Clement and Pimentel, Marco AF and Rajan, Ronnie and Khan, Shadab},
    title = {BioToken and BioFM - Biologically-Informed Tokenization Enables Accurate and Efficient Genomic Foundation Models},
    elocation-id = {2025.03.27.645711},
    year = {2025},
    doi = {10.1101/2025.03.27.645711},
    publisher = {Cold Spring Harbor Laboratory},
    URL = {https://www.biorxiv.org/content/early/2025/04/01/2025.03.27.645711},
    eprint = {https://www.biorxiv.org/content/early/2025/04/01/2025.03.27.645711.full.pdf},
    journal = {bioRxiv}
}
```