Synthetic benchmark datasets and derived results for Tosa

This dataset contains the synthetic sequencing data, reference files, ground-truth count tables, and derived benchmark results used to evaluate Tosa, a high-performance tool for splice-junction (SJ) and exon–intron boundary (EIB) quantification from BAM and CRAM files. The data accompany the manuscript “Tosa: a unified framework for splice-junction and exon-intron boundary quantification reveals methodological differences among RNA-seq counting tools.” All datasets were generated computationally with fixed random seeds and contain no human participants, biological specimens, or personally identifiable information. Dataset contents Paired-end splice-junction validation datasetA minimal paired-end RNA-seq dataset designed to test fragment-aware SJ counting. It contains read pairs in which both mates span the same junction, only one mate spans the junction, or neither mate spans a junction. The archive includes the synthetic reference genome, annotation, paired FASTQ files, and expected junction counts. Bulk exon–intron boundary benchmarkA synthetic genome containing 200 genes with variable exon counts (2, 3, 5, or 10 exons), exon lengths (100–2,000 bp), and intron lengths (100–100,000 bp). Reads were generated to span annotated exon–intron boundaries at predefined coverage levels of 5, 20, or 100 reads per boundary. Both paired-end and single-end libraries are provided, together with gene and read metadata and analytically derived ground-truth counts for boundary-anchor lengths of 1, 4, and 8 bp. Single-cell splice-junction benchmarkA synthetic 10x Genomics Chromium v3-style RNA-seq dataset containing 3,000 cell barcodes and 20 intron-containing genes. Per-cell junction coverage was sampled from 0, 1, 3, or 10 molecules. The archive includes the synthetic genome, GTF annotation, barcode whitelist, paired FASTQ files containing cell barcodes and UMIs, and the ground-truth barcode–junction count table. These data were used to evaluate SJ detection and UMI-deduplicated counting by Tosa, STARsolo, and regtools. Single-cell exon–intron boundary benchmarkA synthetic 10x Genomics Chromium v3-style dataset containing 3,000 cells and reads that continuously span annotated exon–intron boundaries. Per-cell boundary coverage was sampled from 1, 5, or 20 molecules. The dataset uses the same synthetic reference genome as the bulk EIB benchmark and includes a barcode whitelist, paired FASTQ files, and ground-truth barcode–boundary count tables for boundary-anchor lengths of 1, 4, and 8 bp. Derived results The deposit also contains machine-readable accuracy summaries, per-cell accuracy tables, normalized count tables, and Snakemake benchmark records used to generate the corresponding results and figures in the manuscript. Benchmark records report wall-clock time and peak resident set size for the evaluated thread and parameter settings. File organization tosa_synthetic_pe_validation_v1.0.tar.gz: minimal paired-end SJ validation data; tosa_synthetic_bulk_eib_v1.0.tar.gz: bulk EIB reference, reads, metadata, and ground truth; tosa_synthetic_scrna_sj_v1.0.tar.gz: single-cell SJ reference, reads, whitelist, and ground truth; tosa_synthetic_scrna_eib_v1.0.tar.gz: single-cell EIB reads, whitelist, and anchor-specific ground truth; MANIFEST.tsv: file descriptions and provenance; Reproducibility The datasets were generated using the Snakemake workflow available from the Tosa benchmark repository. Generation parameters and random seeds are recorded in the accompanying configuration files. The synthetic reference genomes, annotations, FASTQ files, barcode whitelists, and ground-truth tables are provided so that alignment, quantification, and accuracy evaluation can be reproduced without external biological data. STAR genome indices and alignment BAM files are not included because they can be regenerated from the deposited reference and FASTQ files. Tosa source code is available at https://github.com/Sika-Zheng-Lab/Tosa . The benchmark workflow is available at https://github.com/Sika-Zheng-Lab/Tosa_benchmark_2026. Exact software versions and container images are documented in the benchmark repository and the accompanying README.