Paean: A Unified and Efficient Transcriptome Quantification System Using Heterogeneous Computing

The rapid accumulation of RNA-seq data has raised a higher requirement on the performance of analytic methods. However, existing data analysis pipelines, which are all developed on CPU architecture, take tremendous computational resource to analyze gene expression and alternative splicing in two different computation models, limiting the applications of large RNA-seq datasets. We developed a parallel computing system on GPU-CPU platforms (Parallel analyzer, Paean), which integrates two efficient and highly parallel algorithms into one unified pipeline to facilitate rapid calculation of expression levels and alternative splicing ratios of all genes. With comparable quantification precision to other methods, Paean requires significantly less computation resource but can quantify transcriptome by >100-fold faster than the existing methods. Paean has been adapted to a variety of computational platforms, and its accuracy, reliability and applicability in big datasets were further validated. Large-scale applications including analysis of the complete TCGA dataset (10,866 samples) completed within 3400 min demonstrate excellent performance on massive datasets. Furthermore, Paean enables rapid identification of tumor neoantigen candidates derived from RNA splicing events, providing critical applications for cancer immunotherapy and precision medicine. Paean is open-source and its code repository is available at https://github.com/Bio-Acc/Paean.