Identifying and ranking potential cancer drivers using representation learning on attributed network

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Cancer development is driven by accumulating genomic alterations, with only a subset of mutations acting as drivers.
• Accurate identification of cancer driver genes is critical for advancing drug design, diagnostics, and treatment strategies.

Purpose of the Study:

• To introduce a novel method, Representation Learning on Attributed Graphs (RLAG), for identifying potential cancer driver genes.
• To leverage both network structure and node attributes for gene feature representation in cancer driver prediction.

Main Methods:

• Developed RLAG, a novel approach integrating network topology and gene attributes for feature learning.
• Applied feature vectors to subgroup genes and ranked potential drivers based on intrinsic properties and subgroup importance.
• Validated the method on lung, breast, and prostate cancer datasets.

Main Results:

• The RLAG method demonstrated superior performance compared to three state-of-the-art methods.
• Achieved higher Precision, Recall, and F1-score values in predicting cancer driver genes.
• Successfully identified potential driver genes across multiple cancer types.

Conclusions:

• RLAG offers a robust and effective approach for identifying cancer driver genes.
• The method's ability to integrate network and attribute data enhances prediction accuracy.
• This advancement holds promise for improving cancer diagnostics and therapeutic development.