GLOF: A large-scale expert-curated benchmark of missense variant functional effects

GLOF (Gain and Loss Of Function) is a benchmark dataset of 112,399 missense variants across 2,809 human genes, each classified as LOF (loss-of-function), GOF (gain-of-function), or Neutral by board-certified clinical geneticists.

Dataset Description

The dataset was curated at Mendelics Analise Genomica, one of Latin America's largest clinical genomics laboratories. The annotation process integrated ClinVar pathogenicity classifications, published functional studies, established gene-disease relationships, and expert clinical judgment following ACMG/AMP guidelines.

• Pathogenic variants were sourced from ClinVar (July 2023 release) and cross-referenced against the March 2026 release; variants with reclassified or conflicting evidence were excluded.
• Neutral variants were drawn from gnomAD v3.1 and validated against v4.1 allele frequencies, selecting missense variants with AF > 1%.

Dataset Schema

Field	Type	Description
VARIANTKEY	String	Unique variant identifier: chr-position-ref-alt (GRCh38)
LABEL	String	Functional classification: Neutral, LOF, or GOF
ENSG	String	Ensembl gene identifier
GENE_SYMBOL	String	HGNC gene symbol
AA_POSITION	Integer	Amino acid substitution position in the canonical protein
PROTEIN_REF	Character	Reference (wild-type) amino acid
PROTEIN_ALT	Character	Alternate (mutant) amino acid

Class Distribution

Class	Variants	Percentage	Genes
Neutral	83,902	74.6%	2,749
LOF	25,368	22.6%	2,020
GOF	3,129	2.8%	260

Usage

from datasets import load_dataset

dataset = load_dataset("victormaricato/glof")

Citation

If you use this dataset, please cite:

@article{maricato2026glof,
  title={GLOF: A large-scale expert-curated benchmark dataset of gain-of-function and loss-of-function missense variants},
  author={Maricato, Victor and Schlesinger, David and de Souza Moura, Pedro Nuno},
  year={2026}
}

License

This dataset is released under the CC BY 4.0 license.