Nucleotide Transformer benchmark
Eighteen downstream tasks are benchmark instances, not eighteen unique task families.
67 benchmark-wide protocols · 821 benchmark-wide rows (not a task subtotal)D1 · Sequence understanding & regulation
Chromatin accessibility prediction
Predict cell-type-specific accessible chromatin from DNA sequence.
Task definition
- Input granularity
- long sequence/window
- Biological target
- ATAC/DNase accessibility
- Output representation
- continuous track or binary label
- ML formulation
- regression/classification
- Counting notes
- NR
Benchmark coverage
4 benchmarks cover this task
BEND
Human-genome tasks with biologically aligned splits; inspect each track for leakage controls.
7 benchmark-wide protocols · 105 benchmark-wide rows (not a task subtotal)DART-Eval
Regulatory DNA evaluation across zero-shot, probing and fine-tuning regimes.
0 benchmark-wide protocols · 0 benchmark-wide rows (not a task subtotal)DNALongBench
Long-range genomic tasks; exact sequence length and cell-type packaging vary by track.
2 benchmark-wide protocols · 6 benchmark-wide rows (not a task subtotal)Evaluation protocols
1 protocol units
| Protocol | Track | Split / subset | Comparability |
|---|---|---|---|
PROT-B03-CHROMATIN-ACCESSIBILITY-OFFICIAL-1--AA381D79-8F6D5C54 | chromatin_accessibility B03-TR02 | official 1,354,042/279,422/372,153 split chromatin accessibility | same_table_comparable inference_budget |