Using association signal annotations to boost similarity network fusion

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data integration offers powerful opportunities for complex problem-solving, including disease subtyping and mapping.
• Similarity Network Fusion (SNF) is a promising method for identifying disease subtypes by integrating diverse omics data.
• SNF effectively fuses patient similarity networks from different omics data types, leveraging shared and complementary information.

Purpose of the Study:

• To introduce a novel data integration method, association-signal-annotation boosted similarity network fusion (ab-SNF), to enhance disease subtyping.
• To improve the performance of SNF by incorporating feature-level association signal annotations as weights.
• To identify more accurate and biologically meaningful disease subtypes using multi-omics data.

Main Methods:

• Developed the ab-SNF method, which assigns weights to features based on association signal annotations.
• Constructed subject similarity networks by up-weighting signal features and down-weighting noise features.
• Applied the ab-SNF method to somatic mutation, DNA methylation, and gene expression data from The Cancer Genome Atlas (TCGA).

Main Results:

• The ab-SNF method significantly outperforms the original SNF approach in simulation studies.
• The performance improvement in subtyping is amplified through the integration process using ab-SNF.
• Application to TCGA cancer data revealed new subtypes that more accurately predict patient survival and possess greater biological relevance.

Conclusions:

• The ab-SNF method provides a robust framework for multi-omics data integration and disease subtyping.
• Feature-level annotation weighting enhances the identification of clinically relevant and biologically meaningful subtypes.
• The ab-SNF method offers a valuable tool for advancing precision medicine through improved cancer subtyping.