satMutMPRA

Predict signed reporter-activity effects for variants in saturation-mutagenesis MPRAs using the full assayed regulatory-element sequence in reporter-construct orientation and its assay context.

Task design

published element panels, each containing 50 variants sampled as five candidates from each of ten measured-effect quantiles. Every panel uses opaque candidate IDs and contains no measured effect or selection label in its prompt.

The primary score is mean within-element Spearman correlation. Mean Pearson correlation reports numerical agreement, valid-output rate reports strict JSON compliance, and invalid completed outputs contribute zero while remaining identifiable as format failures.

Task version: 1.1 · question schema 2.0
Output: FINAL: {"V01": number, ...}
Questions: Public development set

Interpretation

Spearman measures ordering within the sampled panel, not absolute calibration. Reporter effects are specific to an assay construct, cell line, and experimental condition and need not transfer to native chromatin or clinical phenotype. Quantile-balanced sampling broadens effect coverage but does not reproduce the natural effect distribution.

Questions

Inspect the exact prompt given to a model alongside its complete response.

questions match the current filters · select one row to inspect it