Hybrid deep learning method to identify key genes in autism spectrum disorder

Area of Science:

• Genetics
• Computational Biology
• Neuroscience

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental condition with significant genetic underpinnings.
• Identifying key genes is crucial for understanding ASD etiology and developing targeted interventions.
• Existing methods for gene identification may not fully capture the intricate genetic interactions in ASD.

Purpose of the Study:

• To introduce a hybrid deep learning approach for identifying key regulator genes in autism spectrum disorder (ASD).
• To evaluate the effectiveness of the proposed method in pinpointing ASD-associated genes.
• To provide a robust framework for ASD research and the development of therapeutic systems.

Main Methods:

• Construction and analysis of a protein-protein interaction network using a graph convolutional network (GCN).
• Feature extraction from gene interactions via GCN, followed by logistic regression for gene prediction.
• Evaluation of identified genes using a susceptible-infected (SI) model and comparison with established databases (SFARI, EAGLE).

Main Results:

• The hybrid deep learning method demonstrated superior performance in identifying key ASD-associated genes compared to traditional centrality methods.
• The susceptible-infected (SI) model confirmed higher "infection ability" for genes identified by the proposed approach.
• Cross-validation with SFARI and EAGLE frameworks confirmed the strong association of identified genes with ASD.

Conclusions:

• The proposed hybrid deep learning method is effective and robust for identifying key regulator genes in autism spectrum disorder.
• This approach offers significant potential for advancing ASD diagnostics, therapeutic strategies, and neural engineering.
• The findings reinforce the genetic complexity of ASD and provide a valuable tool for future research.