MiRNA-gene network embedding for predicting cancer driver genes.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cancer development is driven by mutations in specific genes.
• Accurate identification of cancer driver genes is crucial for targeted therapies and drug design.
• Existing computational methods often overlook the regulatory role of microRNAs (miRNAs) in gene expression and disease.

Purpose of the Study:

• To develop a novel computational approach, GM-GCN, for identifying cancer driver genes.
• To leverage gene-miRNA interaction networks for improved driver gene prediction.
• To integrate miRNA regulatory information into cancer driver gene identification.

Main Methods:

• Construction of a comprehensive gene-miRNA network incorporating regulatory relationships.
• Application of a graph convolution network (GCN) to learn gene feature representations from the network.
• Utilizing a 1D convolution module for feature dimensionality adjustment.
• Employing logistic regression with learned and original gene features for driver gene prediction.

Main Results:

• GM-GCN demonstrated superior performance in predicting cancer driver genes compared to state-of-the-art methods.
• The model achieved high accuracy, as indicated by the area under the receiver operating characteristic curve and the area under the precision-recall curve.
• The approach proved effective for both pan-cancer and individual cancer type predictions.

Conclusions:

• The proposed GM-GCN model effectively identifies cancer driver genes by incorporating gene-miRNA interactions.
• This method offers a significant advancement over existing network-based approaches that do not consider miRNA regulation.
• GM-GCN provides a valuable tool for cancer research, potentially aiding in the development of new diagnostic and therapeutic strategies.