Skip to content

Dataset · songlab

TraitGym DNA model and genetic mutation benchmark

songlab/TraitGym

Dataset for evaluating genomic language models on the task of finding causal regulatory variants in human genetics.

What's inside

Contains tabular and text data ranging from 10 to 100 million records, including a test split on Mendelian traits (mendelian_traits). The dataset is designed to test the ability of genomic models (gLM) to predict the impact of genetic variants. The card also provides a link to the leaderboard and a ready-to-use notebook for quick evaluation in Google Colab.

What you can do with it

  • Evaluating genomic language model (gLM) performance in human genetics tasks
  • Benchmarking DNA regulatory variant effect prediction
  • Comparing bioinformatics algorithms using a standardized leaderboard
Size
10M–100M rows
Downloads this month
22.5K
Likes
14

How to load it

from datasets import load_dataset

ds = load_dataset("songlab/TraitGym")