Identifying Cancer-Related lncRNAs Based on a Convolutional Neural Network

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• Accurate early cancer diagnosis and effective treatment remain significant challenges.
• Long non-coding RNAs (lncRNAs) are increasingly recognized for their crucial roles in disease pathogenesis, particularly in cancers, by regulating gene expression.
• Identifying cancer-specific lncRNAs is vital for understanding cancer mechanisms and developing novel therapeutic strategies.

Purpose of the Study:

• To develop and validate a computational method for identifying cancer-related lncRNAs using machine learning.
• To leverage lncRNA target genes and tissue expression specificity as key features for prediction.
• To improve the identification of lncRNAs associated with various cancer types.

Main Methods:

• Application of Convolutional Neural Networks (CNNs) for predicting cancer-related lncRNAs.
• Utilizing lncRNA target genes and their tissue expression specificity as input features.
• Employing Deep Belief Networks (DBNs) for unsupervised feature encoding of lncRNAs.
• Building individual CNN models for each of the 41 cancer types studied.

Main Results:

• Identification of numerous lncRNAs associated with 41 distinct cancer types.
• Demonstrated superior performance compared to existing methods, achieving an Area Under the Curve (AUC) of 0.81 and Area Under the Precision-Recall Curve (AUPR) of 0.79.
• Validation of the method's accuracy through rigorous ten-cross validation and case studies.

Conclusions:

• The proposed CNN-based computational method effectively identifies cancer-related lncRNAs.
• This approach offers a promising tool for advancing cancer research, diagnosis, and treatment development.
• The findings highlight the potential of lncRNAs as biomarkers and therapeutic targets in oncology.