MODILM: towards better complex diseases classification using a novel multi-omics data integration learning model.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate classification of complex diseases is vital for personalized medicine.
• Multi-omics data integration enhances disease analysis due to correlated, comprehensive, and complementary information.
• Challenges in multi-omics integration include data imbalance, scale variation, heterogeneity, and noise.

Purpose of the Study:

• To propose a novel multi-omics data learning model (MODILM) for improved complex disease classification.
• To extract significant and complementary information from diverse omics data for enhanced accuracy.
• To provide a tool supporting clinical diagnosis decision-making.

Main Methods:

• Constructing similarity networks for each omics data using cosine similarity.
• Employing Graph Attention Networks to learn intra-association features from single-omics data.
• Utilizing Multilayer Perceptron networks for feature mapping and strengthening.
• Fusing features with a View Correlation Discovery Network for cross-omics feature learning.

Main Results:

• MODILM demonstrated superior performance on six benchmark datasets (miRNA, mRNA, DNA methylation).
• The model effectively improved complex disease classification accuracy compared to state-of-the-art methods.
• Achieved unique class-level distinctiveness through fused cross-omics features.

Conclusions:

• MODILM offers a competitive approach for extracting and integrating multi-omics data.
• The model effectively captures complementary information for improved disease classification.
• MODILM presents a promising tool for clinical diagnosis support.