Comprehensive assessment of isoform detection methods for third-generation data

The advancement of Third-Generation Sequencing (TGS) techniques managed to increase the sequencing length to several kilobases, which leads to a bright future for completely reserving alternative splicing (AS) events and isoform expressions. In recent years, many computational methods for isoform detection from long-read sequencing data have been developed and published. However, there is no prior comparative study that systemically evaluates the performance of the software implemented with different algorithms. Here we benchmarked nine methods implemented in seven computational tools that can identify isoform structures from TGS RNA sequencing data and analyzed their performances from various aspects using both simulated datasets produced by an in-house simulator and previously published experimental data. Our results comprehensively demonstrate the relative effectiveness of the approaches and provide guidance as well as recommendations for future research on AS analysis and further improvement of the tools for isoform detection using TGS data.

第三代测序（Third-Generation Sequencing, TGS）技术的进步将测序读长提升至数千碱基对，为完整保留可变剪接（alternative splicing, AS）事件与异构体表达信息带来了光明前景。近年来，诸多可从长读长测序数据中检测异构体的计算方法已被开发并发表。然而目前尚无系统性对比研究对不同算法实现的软件工具性能开展全面评估。本研究针对7款可从TGS RNA测序数据中识别异构体结构的计算工具所包含的9种方法开展了基准测试，并利用自研模拟器生成的模拟数据集与已发表的实验数据集，从多维度分析了各方法的性能表现。本研究结果全面阐明了各类方法的相对有效性，可为后续可变剪接分析相关研究以及基于TGS数据的异构体检测工具的进一步优化提供指导与建议。