Flnc: Machine Learning Improves the Identification of Novel Long Noncoding RNAs from Stand-Alone RNA-Seq Data

Area of Science:

• Genomics and transcriptomics
• Molecular biology
• Bioinformatics

Background:

• Long noncoding RNAs (lncRNAs) are crucial regulators in human development and disease.
• Existing RNA sequencing (RNA-seq) data is abundant, yet many lncRNAs remain unannotated.
• Current methods for lncRNA identification have limitations, including high false discovery rates and inability to detect single-exon lncRNAs.

Purpose of the Study:

• To develop a robust computational tool for accurate identification of novel and annotated full-length lncRNAs.
• To overcome the limitations of existing methods in identifying lncRNAs from RNA-seq data.
• To enable the discovery of single-exon lncRNAs without requiring additional experimental data.

Main Methods:

• Development of Flnc software integrating machine learning models.
• Incorporation of features such as transcript length, promoter signature, exon number, and genomic location.
• Direct analysis of RNA sequencing data without reliance on transcriptional initiation profiling.

Main Results:

• Flnc achieves state-of-the-art prediction performance with an AUROC score exceeding 0.92.
• Significantly improves prediction accuracy from under 50% to over 85% compared to conventional methods.
• Successfully identifies both novel and annotated full-length lncRNAs, including single-exon variants.

Conclusions:

• Flnc provides an accurate and efficient method for lncRNA identification from RNA-seq data.
• The software addresses key limitations of previous approaches, enhancing lncRNA discovery.
• Flnc is readily available for use by the research community via GitHub.