Genome Understanding Evaluation (GUE)
28 datasets/tasks in the native packaging; mapped here to five task families.
1 benchmark-wide protocols · 10 benchmark-wide rows (not a task subtotal)D1 · Sequence understanding & regulation
TF binding-site prediction
Predict sequence-specific TF occupancy or binding sites.
Task definition
- Input granularity
- sequence/window or base
- Biological target
- TF occupancy/site
- Output representation
- probability or per-base score
- ML formulation
- classification/segmentation
- Counting notes
- NR
Benchmark coverage
7 benchmarks cover this task
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)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)Benchmarking DNA foundation models for genomic and genetic tasks
One-time benchmark study rather than a maintained benchmark suite.
6 benchmark-wide protocols · 8 benchmark-wide rows (not a task subtotal)ENCODE-DREAM in vivo TF binding
Held-out cell types and chromosomes improve realism, but access is through Synapse.
0 benchmark-wide protocols · 0 benchmark-wide rows (not a task subtotal)OmniGenBench
Meta-suite packaging RGB, BEACON, GUE, Genomic Benchmarks and PGB; native_task_count=5 suites, not the sum of their downstream tasks.
0 benchmark-wide protocols · 0 benchmark-wide rows (not a task subtotal)Evaluation protocols
0 protocol units
This task has benchmark mappings but no reconstructable numeric protocol in the current snapshot.