Predicting lncRNA-miRNA Interaction via Graph Convolution Auto-Encoder.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNA (miRNA) and long non-coding RNA (lncRNA) interactions are crucial for gene regulation.
• Existing computational tools for predicting lncRNA-miRNA interactions are limited, and known interactions are scarce.
• lncRNAs and miRNAs exhibit shared patterns, suggesting prediction is feasible via semi-supervised learning.

Purpose of the Study:

• To develop an effective computational model for predicting novel lncRNA-miRNA interactions.
• To leverage side information and inherent patterns in lncRNA and miRNA attributes for improved prediction accuracy.
• To address the limitations of existing prediction tools and the scarcity of known interactions.

Main Methods:

• Proposed GCLMI (Graph Convolution for novel lncRNA-miRNA Interactions), an end-to-end prediction model.
• Combined graph convolution and auto-encoder techniques to integrate raw node attributes and interaction network topology.
• No preprocessing of feature information was required, allowing direct incorporation of data.

Main Results:

• Experimental results on a real dataset demonstrated the robustness and effectiveness of the GCLMI model.
• K-fold cross-validation confirmed the prediction performance.
• The designed graph convolution layer successfully integrated input data by filtering graph nodes based on features.

Conclusions:

• GCLMI is a robust and effective computational tool for predicting potential lncRNA-miRNA interactions.
• The model can utilize various numerical features describing lncRNAs or miRNAs provided by users.
• GCLMI is anticipated to significantly contribute to identifying novel lncRNA-miRNA interactions for gene regulation studies.