LIMO-GCN: a linear model-integrated graph convolutional network for predicting Alzheimer disease genes

Area of Science:

• Computational biology
• Genetics
• Neuroscience

Background:

• Alzheimer's disease (AD) genetic etiology remains incompletely understood.
• Gene network analysis shows promise for predicting AD-associated genes.
• Existing methods struggle to model complex linear and nonlinear relationships in gene networks.

Purpose of the Study:

• To develop a novel computational method for predicting Alzheimer's disease genes.
• To address limitations in existing gene network-based prediction models.
• To improve the accuracy of identifying genes linked to Alzheimer's disease.

Main Methods:

• Proposed Linear Model-integrated Graph Convolutional Network (LIMO-GCN).
• Integrated a linear model with Graph Convolutional Network (GCN) to capture both linear and nonlinear data patterns.
• Applied LIMO-GCN to predict Alzheimer's disease genes using network data.

Main Results:

• LIMO-GCN demonstrated superior performance compared to state-of-the-art methods like GCN, network-wide association studies, and random walk.
• Top-ranked genes predicted by LIMO-GCN showed significant association with AD, supported by molecular evidence.
• The method effectively models both linearity and nonlinearity in gene network data.

Conclusions:

• LIMO-GCN offers a novel and effective approach for prioritizing Alzheimer's disease genes.
• The integration of linear models with GCN enhances gene prediction accuracy.
• This method advances the understanding of AD's genetic underpinnings.