Pseudo2GO: A Graph-Based Deep Learning Method for Pseudogene Function Prediction by Borrowing Information From Coding...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Pseudogenes are increasingly recognized for their functional roles, moving beyond their historical classification as evolutionary relics.
• Predicting pseudogene functions using computational methods, specifically for Gene Ontology (GO) terms, is crucial for guiding experimental research.
• Existing computational methods lack pseudogene specificity and struggle with limited features and functional annotations, hindering predictive model development.

Purpose of the Study:

• To develop a novel computational method for predicting Gene Ontology (GO) terms for pseudogenes.
• To leverage the functional similarity between pseudogenes and their parent coding genes by borrowing information.
• To establish a graph-based deep learning approach for enhanced pseudogene function prediction.

Main Methods:

• Proposed Pseudo2GO, a graph-based deep learning semi-supervised model for pseudogene function prediction.
• Constructed a sequence similarity graph connecting pseudogenes and coding genes.
• Incorporated diverse features including expression profiles, microRNA interactions, protein-protein interactions (PPIs), and genetic interactions as node attributes.
• Utilized graph convolutional networks for attribute propagation and pseudogene classification.

Main Results:

• Pseudo2GO demonstrated state-of-the-art performance in pseudogene function prediction.
• The method significantly outperformed existing frameworks adapted from protein function prediction.
• Superior performance was particularly evident in the M-AUPR (Maximal Area Under the Precision-Recall Curve) metric.

Conclusions:

• Pseudo2GO offers an effective solution for predicting pseudogene functions, addressing the limitations of previous methods.
• The graph-based deep learning approach successfully utilizes information from coding genes to infer pseudogene functions.
• This method advances the field of pseudogene functional genomics and aids in prioritizing experimental validation.