Using synthetic RNA to benchmark poly(A) length inference from direct RNA sequencing.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Polyadenylation is crucial for RNA regulation, affecting mRNA decay, translation, and isoform specificity.
• Direct RNA sequencing offers full-length RNA analysis, enabling transcriptome-wide poly(A) tail studies.
• Existing poly(A) tail estimation tools lack comprehensive benchmarking against gold-standard datasets.

Purpose of the Study:

• To introduce BoostNano, a novel deep learning tool for poly(A) tail length estimation.
• To benchmark BoostNano against established tools (tailfindr, nanopolish) and a deep learning tool (Dorado).
• To evaluate tool performance using synthetic RNA standards with known poly(A) tail lengths.

Main Methods:

• Development of the BoostNano deep learning model for poly(A) estimation.
• Evaluation of four tools (BoostNano, tailfindr, nanopolish, Dorado) on synthetic RNA datasets (Sequin, eGFP).
• Analysis of tool accuracy based on known ground-truth poly(A) tail lengths.

Main Results:

• Tool performance varied depending on poly(A) tail length and sample type.
• Averaging poly(A) estimates over multiple reads improved accuracy.
• Dorado demonstrated superior performance with fast runtimes, low mean error, and ease of use.

Conclusions:

• Dorado is recommended for poly(A) tail length estimation in direct RNA sequencing.
• Accurate poly(A) tail analysis is vital for understanding transcript stability and regulation.
• This benchmark provides a reference for improving transcriptome analysis and RNA regulatory studies.