Card deck

Nucleotide Transformer: one DNA model, 18 genomics tasks

The Nucleotide Transformer is a BERT-style foundation model for DNA, pretrained on thousands of genomes with no labels, then benchmarked on 18 tasks. It matches specialist tools, learns genomic elements on its own, and in version 2 a 250-million-parameter model beat the 2.5-billion one at a tenth of the size.

Genomics × AIPaper

1 / 12 Nucleotide Transformer: one DNA model, 18 genomics tasks

A twelve-slide breakdown of the Nucleotide Transformer, the origin of the NT line: how a masked language model reads DNA in 6-mer tokens, what it learns about introns, coding regions and variants with no supervision, why it matched SpliceAI on splice-site prediction, and how version 2 reached the top benchmark score with a model ten times smaller than the 2.5-billion one.

Read the paper: Nucleotide Transformer: building and evaluating robust foundation models for human genomics (Nature Methods, 2025).