Deep learning based DNA:RNA triplex forming potential prediction.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Long non-coding RNAs (lncRNAs) can regulate gene expression by forming DNA:RNA triplex structures.
• Current prediction methods rely on statistical rules, leading to many false positives and lacking experimental validation.
• Existing approaches fail to incorporate features from experimentally verified triplex-forming lncRNAs.

Purpose of the Study:

• To develop a novel machine learning model for accurate prediction of DNA:RNA triplex formation.
• To identify lncRNAs with high potential for triplex formation and predict target DNA sites.
• To overcome limitations of current statistical methods in triplex prediction.

Main Methods:

• Developed TriplexFPP, the first machine learning model for DNA:RNA triplex prediction.
• Utilized convolutional neural networks to learn high-level features from experimentally verified data.
• Employed fivefold cross-validation for performance assessment.

Main Results:

• TriplexFPP achieved high accuracy in predicting triplex-forming lncRNAs (AUC=0.9649, PRC=0.9996).
• The model also demonstrated strong performance in predicting triplex DNA sites (AUC=0.8705, PRC=0.9671).
• Summarized cis and trans targeting mechanisms of triplex-forming lncRNAs.

Conclusions:

• TriplexFPP effectively predicts lncRNAs capable of forming triplexes and their potential DNA targets.
• The model offers a significant improvement over existing methods by integrating experimental data.
• This tool may advance the understanding of lncRNA functions through triplex formation.