Deep learning and direct sequencing of labeled RNA captures transcriptome dynamics.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Eukaryotic genes generate diverse RNA isoforms with unique functions.
• Understanding RNA metabolism, including transcription and decay, is crucial for gene regulation.
• Current methods struggle to differentiate and quantify individual RNA isoforms.

Purpose of the Study:

• To introduce RNAkinet, a novel deep learning approach for RNA isoform detection and kinetic analysis.
• To enable the differentiation of nascent RNA molecules from pre-existing ones.
• To facilitate the quantification of RNA isoform half-lives.

Main Methods:

• Utilized a deep convolutional and recurrent neural network (RNAkinet).
• Employed metabolic labeling with 5-ethynyl uridine.
• Integrated long-read, direct RNA sequencing with nanopore technology.
• Processed raw electrical signals from nanopore sequencing.

Main Results:

• RNAkinet successfully detects nascent RNA molecules.
• The method distinguishes between newly synthesized and older RNA.
• RNAkinet demonstrates generalizable prediction performance across different cell types and organisms.
• Quantification of RNA isoform half-lives was achieved.

Conclusions:

• RNAkinet provides a robust method for analyzing RNA isoform metabolism.
• The tool enables the identification of kinetic parameters for individual RNA isoforms.
• Facilitates deeper insights into gene regulation and RNA processing.