NVAR

NVAR is an evaluation-focused benchmark for studying mutation-aware reasoning under natural variants in protein sequences.

The dataset contains:

• reference protein sequences (uniprot_id, reference_*)
• variant protein sequences (variant_*)
• natural language functional descriptions (reference_function)
• structured mutation effect annotations (variant_annotation, effect, target)

NVAR supports evaluation of protein caption generation, variant effect prediction, and mutation-aware reasoning under sequence perturbations.

Column Descriptions

variant_caption: (Deprecated) This column is redundant and will be removed in the final version. Please refer to variant_raw_annotation instead.