OFGPMA: Optimal frequency graph representation learning for pseudogene and miRNA association prediction

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Pseudogenes and microRNAs (miRNAs) form competitive endogenous RNA (ceRNA) networks with mRNAs, regulating cellular mechanisms.
• Dysregulation of these networks is linked to various pathological conditions, highlighting their diagnostic potential.
• Current methods for identifying pseudogene-miRNA associations (PMAs) are experimental, time-consuming, and costly.

Purpose of the Study:

• To develop an effective computational framework for predicting novel pseudogene-miRNA associations (PMAs).
• To address the limitations of existing experimental approaches for PMA identification.

Main Methods:

• Proposed OFGPMA, an optimal frequency graph representation learning framework for PMA prediction.
• Enhanced graph neural network expressiveness using Rayleigh and Chebyshev pooling to learn high- and low-frequency energy components.
• Integrated global graph topology via Random Walk with Restart (RWR) and local substructure features via enclosing subgraph analysis.

Main Results:

• OFGPMA demonstrated superior performance compared to state-of-the-art methods in pseudogene-miRNA association prediction.
• The framework exhibited excellent generalization capabilities in comprehensive experiments.

Conclusions:

• OFGPMA offers an efficient and effective computational approach for identifying PMAs.
• This method holds promise for advancing disease diagnostics through better understanding of ceRNA networks.