View source

Tasks

VEP-bench publishes transparent development tasks with exact model prompts, complete responses, and deterministic scores.

Fitness (SGE)

Predict continuous functional effects for panels sampled from endogenous-locus saturation genome editing assays.

  • One eligible exon-level question per gene with 50 complete assayed alleles
  • Gene, assay mechanism, and exon sequence with exact 100 bp flanks
  • Mean within-gene Spearman correlation, with Pearson and output-validity diagnostics
  • Source: published SGE score sets in MaveDB, cited as Rubin et al. (2025)

Open task →

Fitness (SGE): variants in a native exon with 100 bp flanks are measured in a cellular assay. Higher functional-effect scores indicate greater impairment within one gene and assay.

Expression (satMutMPRA)

Predict signed reporter-activity effects for panels sampled across the measured effect distribution of 16 regulatory elements.

  • 16 element-level questions with 50 variants each
  • Full assayed element sequence, assay context, and compact VCF input
  • Mean within-element Spearman correlation, with Pearson and output-validity diagnostics
  • Source: Kircher et al. (2019), distributed in the CADD v1.7 RegSeq collection

Open task →

Expression (satMutMPRA): promoter or enhancer inserts drive a luciferase reporter. Predict signed log₂ changes in reporter activity relative to the reference.

Splicing (OpenSplice)

Predict signed changes in exon inclusion for score-space-sampled allele panels in complete source-derived three-exon minigene cassettes.

  • 20 exon-level questions from 20 distinct genes, with 50 complete alleles each
  • Complete cassette sequence, exact segment intervals, and HEK293T assay context
  • Mean within-exon Spearman correlation, with Pearson and output-validity diagnostics
  • Source: Quarantani et al. (2026) and the OpenSplice Figshare v5 dataset

Open task →

Splicing (OpenSplice): variants in a three-exon cassette can change inclusion or skipping of the middle exon. Predict ΔPSI, the change in exon inclusion relative to the reference.