Inside Genebench-Pro

OpenAI has published ten case studies showcasing representative questions from Genebench-Pro, providing original prompts, datasets, and supporting materials for each scenario. An announcement blog is available for an overview of the benchmark and key findings. The case studies cover a diverse range of complex genomic and clinical data challenges. Representative tasks include estimating the clinical utility of a synthetic TXR1-directed inhibitor in tumors driven by structural variants, determining whether apparent lncRNA dependencies are transcript-specific or driven by nearby loci, and estimating direct disease effects for nearby proteins using cis multivariable Mendelian randomization. Additional case studies focus on carrier-screening assay data to estimate ancestry-specific carrier frequencies and residual risks, evaluating genotype effects on activated-monocyte expression after removing ambient RNA from single-cell RNA-seq data, and assessing calibrated clinical associations for nested structural subhaplotypes inside anonymous inversion-like loci. Further scenarios involve quantifying focal case-control Hi-C loop-strength differences by removing mapping and structural variant artifacts, mapping a chromosome-1 quantitative-trait locus in an eight-founder recombinant population via founder ancestry reconstruction, and inferring parent-specific ancestry proportions and recent admixture timing from phased local-ancestry tracts. Finally, the benchmark includes tasks to infer which of two haploid loci is under stronger positive selection from ancient allele-frequency time series while accounting for error, drift, and changing population size.