DeepLGP: a novel deep learning method for prioritizing lncRNA target genes.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Long non-coding RNAs (lncRNAs) are crucial regulators of protein-coding genes (PCGs) and implicated in complex diseases.
• The rapid increase in lncRNA data necessitates computational methods for predicting their target PCGs.
• Existing methods lack computational approaches for predicting novel lncRNA targets.

Purpose of the Study:

• To develop a computational method, DeepLGP, for prioritizing target PCGs of lncRNAs.
• To leverage graph convolutional networks (GCNs) for predicting lncRNA-gene interactions.
• To facilitate the identification of disease-associated genes through lncRNA target prediction.

Main Methods:

• Utilized a graph convolutional network (GCN) framework (DeepLGP) for lncRNA target gene prediction.
• Integrated genomic location, expression data across 13 tissues, and miRNA-mediated interactions as features.
• Employed a convolutional neural network for prioritizing target genes based on learned features.

Main Results:

• DeepLGP achieved high performance in 10-cross validations, with AUCs ranging from 0.90-0.98 and aUPRCs from 0.91-0.98.
• Identified that highly similar lncRNAs share a significant number of overlapped target genes.
• Demonstrated that genes co-targeted by lncRNA sets are likely involved in the same diseases.

Conclusions:

• DeepLGP is an effective computational tool for predicting lncRNA target genes.
• The findings support the role of lncRNA-PCG interactions in disease mechanisms.
• This approach can aid in discovering novel disease-causing PCGs.