RloopBench: curated R-loop benchmark datasets across experimental platforms and species

This dataset accompanies RloopBench, a benchmark for evaluating genomic sequence representations for R-loop prediction across experimental platforms and species. It contains curated R-loop-positive sequences and corresponding negative controls, organized into four directories: DeepER_RChIP: Human R-ChIP benchmark data originally used in the DeepER study, with training, validation, and test splits. RloopTracker_DRIPc: An independent human DRIPc-seq validation dataset for cross-platform evaluation. RloopBase: Human R-loop regions spanning R-loopBase consensus levels 4–9, with corresponding matched proxy negative controls. CrossSpecies: Curated datasets for Arabidopsis thaliana, Glycine max, Oryza sativa, Zea mays, Drosophila melanogaster, Danio rerio, and Mus musculus, with corresponding matched proxy negative controls. The R-ChIP benchmark retains its original negative controls. For the other datasets, matched proxy negatives were constructed as described in the accompanying manuscript. These controls should not be interpreted as experimentally confirmed absence of R-loop formation. Dataset sources, preprocessing procedures, and evaluation protocols are described in the accompanying manuscript. Code is available at https://github.com/LinResearchGroup-NCSU/RLoopBench.