Logo image
Back
NADG-GAM: Neighbor aggregation-based neurological disease–gene identification via optimal generative adjacency matrix
Journal article   Peer reviewed

NADG-GAM: Neighbor aggregation-based neurological disease–gene identification via optimal generative adjacency matrix

Mengyuan Jin, Ziyi Deng, Yin Zhang, Jia Liu and Fang Hu
Applied soft computing, Vol.171, 112756
03/2025
DOI: https://doi.org/10.1016/j.asoc.2025.112756
Web of Science ID: WOS:001406370800001

Metrics

1 Record Views

Abstract

Flexible feature dimensionality reduction Network representation learning Neurological disease gene identification Optimal generative adjacency matrix strategy Computational Mathematics Neurological Disorders
The identification of disease-related genes is crucial for advancing early diagnosis, treatment, disease prevention, and precision medicine. The high dimensionality of bioinformatics data and the lack of effective entity and connection extraction techniques frequently result in inadequate predictions in disease–gene studies. To address these challenges, this study proposes a neighbor aggregation-based neurological disease–gene identification method, Neighbor Aggregation via an Optimal Generative Adjacency Matrix (NADG-GAM). The proposed approach incorporates a flexible feature dimensionality reduction strategy tailored to diverse feature matrices across varying scenarios. The NADG-GAM algorithm leverages an optimal generative adjacency matrix strategy that employs a self-adaptive mechanism to construct affinity matrices optimized for distinct bioinformatics networks. The proposed algorithm enables low-dimensional node representation learning by combining dimensionality-reduced feature matrices with optimum generative adjacency matrices to aggregate neighbor information. Comprehensive experiments were performed on neurological disease datasets, including comparison analyses, ablation tests, and parameter sensitivity assessments, utilizing six standard evaluation criteria. The results indicate higher performance of our model, including accuracy, scalability, adaptability, etc., across various parameters compared to baseline methods. Specifically, it achieved improvements of at least 1.2%, 7.4%, 7%, 6.7%, and 1.3% in AP, Precision, Recall, F1-Score, and AUC. The candidate gene prediction results of NADG-GAM are significant for understanding disease formation mechanisms and experimental verification, and it is promising for diagnosing and treating neurological diseases. Meanwhile, the framework that this algorithm offers functions as an effective idea for broader disease-related studies.

Details

Logo image