Integrating multi-omics data of childhood asthma using a deep association model

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Childhood asthma presents significant mortality and morbidity challenges.
• Multi-omics data offers potential for discovering collaborative biomarkers and diagnostic models.
• Existing models struggle to capture nonlinear associations in multi-omics data.

Purpose of the Study:

• To propose a novel deep association model (DAM) for analyzing multi-omics data in childhood asthma.
• To develop an efficient framework for capturing nonlinear associations and improving diagnostic model interpretability.
• To identify collaborative biomarkers and construct accurate diagnostic models for childhood asthma.

Main Methods:

• Deep Subspace Reconstruction for data fusion and noise reduction.
• Joint Deep Semi-Negative Matrix Factorization for latent pattern identification and biomarker extraction.
• Deep Orthogonal Canonical Correlation Analysis for feature ranking and nonlinear correlation modeling.

Main Results:

• The DAM achieved a prediction AUC of 0.912 on an independent test dataset, outperforming baseline methods.
• Collaborative biomarkers and relevant pathways were identified at both gene expression and methylation levels.
• The model demonstrated effectiveness in algorithm performance and biological significance.

Conclusions:

• DAM provides an interpretable machine learning approach for multi-omics data analysis.
• The model effectively captures nonlinear associations among samples and biological features.
• DAM facilitates the exploration of biomarker candidates and efficient diagnostic models for complex diseases like childhood asthma.