Predicting the potential human lncRNA-miRNA interactions based on graph convolution network with conditional random...

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Long non-coding RNAs (lncRNAs) and microRNAs (miRNAs) are key regulators in biological processes.
• Understanding lncRNA-miRNA interactions is vital for deciphering gene regulation.
• Experimental methods for identifying these interactions are time-consuming and labor-intensive.

Purpose of the Study:

• To develop an efficient computational method for predicting human lncRNA-miRNA interactions.
• To improve the accuracy and speed of identifying functional ncRNA interactions.
• To provide a valuable tool for researchers studying gene regulation.

Main Methods:

• A novel method, GCNCRF, combining Graph Convolutional Neural Networks (GCN) and Conditional Random Fields (CRF).
• Construction of a heterogeneous network integrating known interactions, similarity networks, and feature matrices.
• Incorporation of an attention mechanism within the CRF layer to prioritize important node features.

Main Results:

• GCNCRF achieved an Area Under the Curve (AUC) of 0.947 in 5-fold cross-validation on the main dataset.
• Demonstrated superior prediction accuracy compared to six existing state-of-the-art methods.
• Successfully predicted human lncRNA-miRNA interactions with high efficacy.

Conclusions:

• GCNCRF offers a highly accurate and efficient computational approach for predicting lncRNA-miRNA interactions.
• The method effectively leverages network structures and node features for improved prediction.
• This tool can significantly accelerate research in non-coding RNA functional genomics.