DA-SRN: Omics data analysis based on the sample network optimization for complex diseases

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics
• Metabolomics
• Transcriptomics

Background:

• Accurate biomarker identification and sample classification remain significant challenges in complex disease research.
• Patient similarity network (PSN) analysis offers a robust framework for analyzing high-dimensional omics data.
• The structural properties of PSNs can indicate the feature space's discriminative power.

Purpose of the Study:

• To introduce a novel omics data analysis method, DA-SRN (Data Analysis-Sample Reference Network), for biomarker discovery and sample categorization.
• To optimize network structure and identify informative features using a genetic algorithm.
• To leverage graph neural networks and the sample reference network for accurate sample labeling.

Main Methods:

• Developed DA-SRN, integrating genetic algorithms for feature selection and network optimization.
• Employed graph neural networks for sample classification based on the optimized reference network and selected features.
• Validated DA-SRN against nine existing methods using genomics, metabolomics, and transcriptomics datasets.

Main Results:

• DA-SRN demonstrated superior performance across multiple metrics, including AUROC and AUPRC, compared to existing methods.
• Identified key metabolites associated with type 2 diabetes (T2D) using metabolomics data.
• Pathway analysis confirmed the relevance of identified metabolites to T2D pathogenesis.

Conclusions:

• DA-SRN effectively extracts valuable biological insights from complex omics data by analyzing inter-sample relationships.
• The method shows significant potential for biomarker identification and sample discrimination in complex diseases.
• DA-SRN advances the application of network-based approaches in precision medicine.