Benchmarking long-read RNA-sequencing analysis tools using in silico mixtures.

Area of Science:

• Genomics
• Bioinformatics
• Transcriptomics

Background:

• Accurate isoform detection and differential expression analysis are crucial for understanding gene regulation.
• Existing long-read sequencing workflows lack standardized benchmark datasets for performance evaluation.

Purpose of the Study:

• To establish a benchmark experiment for evaluating long-read isoform detection and differential expression analysis tools.
• To provide ground-truth data for assessing the performance of various bioinformatics workflows.

Main Methods:

• Utilized two human lung adenocarcinoma cell lines, profiled in triplicate.
• Incorporated synthetic, spliced, spike-in RNAs (sequins) for ground-truth validation.
• Performed deep sequencing on both Illumina short-read and Oxford Nanopore Technologies long-read platforms.
• Generated in silico mixture samples to assess performance without true positives/negatives.

Main Results:

• StringTie2 and bambu demonstrated superior performance in isoform detection compared to other tools.
• DESeq2, edgeR, and limma-voom were identified as the top-performing tools for differential transcript expression analysis.
• No single tool emerged as a clear leader for differential transcript usage analysis, indicating a need for further development.

Conclusions:

• The benchmark experiment provides valuable insights into the performance of current long-read analysis tools.
• Specific tools are recommended for isoform detection and differential expression analysis based on performance.
• Further methodological advancements are required for robust differential transcript usage analysis.