A comprehensive evaluation of long-read de novo transcriptome assembly

Area of Science:

• Genomics
• Bioinformatics
• Transcriptomics

Background:

• De novo transcriptome assembly using long-read sequencing is crucial for non-model organisms lacking reference genomes.
• Established protocols for reference-free long-read assembly and differential expression analysis are lacking.
• This study addresses the need for benchmarking long-read assembly tools.

Purpose of the Study:

• To evaluate and compare the performance of long-read de novo transcriptome assembly tools (RATTLE, RNA-Bloom2, isONform) against a short-read assembler (Trinity).
• To assess the impact of assembly choice on differential gene and transcript expression analysis.
• To provide guidance for selecting optimal strategies for reference-free long-read transcriptome analysis.

Main Methods:

• Evaluation of RATTLE, RNA-Bloom2, and isONform using simulated and real datasets (human, pea).
• Comparison with Trinity (short-read assembler) across various sequencing depths and technologies (ONT cDNA, ONT direct RNA, PacBio).
• Assessment of assembly metrics and downstream differential expression analysis.

Main Results:

• Long reads produce longer transcripts than short reads in reference-free assembly, though reference-guided methods remain superior.
• RNA-Bloom2 combined with Corset demonstrated superior accuracy and computational efficiency among the evaluated de novo pipelines.
• Assembly strategy significantly impacts differential gene and transcript expression detection.

Conclusions:

• RNA-Bloom2 with Corset is recommended for reference-free de novo transcriptome assembly and differential expression analysis when a reference genome is unavailable.
• Long-read sequencing offers advantages for de novo assembly but requires further optimization for accuracy and redundancy reduction.
• The study provides essential guidance for researchers working with non-model organisms.