A deep learning method for lincRNA detection using auto-encoder algorithm.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• RNA sequencing (RNA-seq) aids in discovering novel long non-coding RNAs (lincRNAs).
• Deep learning methods offer a revolution in knowledge-based technologies.
• lincRNAs exhibit regulated expression and conserved sequence motifs, justifying deep learning for detection.

Purpose of the Study:

• To develop and validate deep learning methods for identifying lincRNA transcription sequences.
• To leverage auto-encoder algorithms for enhanced lincRNA detection.
• To explore the potential of deep learning in understanding lincRNA biology.

Main Methods:

• Utilized RNA sequencing data and annotated human genome sequences.
• Developed a two-layer deep neural network incorporating an auto-encoder algorithm.
• Employed various encoding schemes to optimize performance on intergenic DNA sequences.

Main Results:

• The auto-encoder achieved 100% and 92.4% prediction accuracy on transcription sites.
• Deep neural networks outperformed traditional methods like support vector machines.
• Identified a novel transcription site, demonstrating the method's predictive power.

Conclusions:

• Deep learning, particularly auto-encoders, effectively learns features for lincRNA detection from genomic sequences.
• This study represents the first application of deep learning for identifying lincRNA transcription sequences.
• The findings highlight the extensive ability of deep learning in lincRNA prediction.